Patent Grant
9952803
This application generally relates to data storage, and more particularly to techniques used in connection with data storage configuration.
Computer systems may include different resources used by one or more host processors. Resources and host processors in a computer system may be interconnected by one or more communication connections. These resources may include, for example, data storage devices such as those included in the data storage systems manufactured by EMC Corporation. These data storage systems may be coupled to one or more host processors and provide storage services to each host processor. Multiple data storage systems from one or more different vendors may be connected and may provide common data storage for one or more host processors in a computer system.
A host processor may perform a variety of data processing tasks and operations using the data storage system. For example, a host processor may perform basic system I/O operations in connection with data requests, such as data read and write operations.
Host processor systems may store and retrieve data using a storage device containing a plurality of host interface units, disk drives, and disk interface units. Such storage devices are provided, for example, by EMC Corporation of Hopkinton, Mass. The host systems access the storage device through a plurality of channels provided therewith. Host systems provide data and access control information through the channels to the storage device and storage device provides data to the host systems also through the channels. The host systems do not address the disk drives of the storage device directly, but rather, access what appears to the host systems as a plurality of logical disk units, logical devices, or logical volumes (LVs). The logical disk units may or may not correspond to the actual disk drives. Allowing multiple host systems to access the single storage device unit allows the host systems to share data stored therein.
In connection with data storage, a variety of different technologies may be used. Data may be stored, for example, on different types of disk devices and/or flash memory devices. The data storage environment may define multiple storage tiers in which each tier includes physical devices or drives of varying technologies, performance characteristics, and the like. The physical devices of a data storage system, such as a data storage array, may be used to store data for multiple applications.
In accordance with one aspect of the invention is a method for configuring data storage comprising: receiving criteria identifying limits for a plurality of storage tiers for one or more logical devices, said plurality of storage tiers including a first storage tier and one or more other storage tiers, said first storage tier being a higher performance storage tier than said one or more other storage tiers; and performing first processing to determine a first set of data portions of the one or more logical devices to be stored on physical devices of the first storage tier subject to limits included in said criteria, wherein said first processing further includes: determining a plurality of scores for a plurality of data portions of the one or more logical devices, wherein one of said plurality of scores is determined for a different one of the plurality of data portions, each of said plurality of scores being calculated using one or more metrics including a first metric weighted based on an expected I/O size; ranking said plurality of data portions in accordance with said plurality of scores; and selecting said first set of data portions from said plurality of data portions based on said ranking. The method may also include determining that other data portions of said plurality of data portions not included in the first set are stored on physical devices of said one or more other storage tiers. Each data portion in the first set may have an associated score that is higher than other scores associated with others of said plurality of data portions of the one or more logical devices not included in the first set. Each data portion in the first set not already located on a physical device of the first storage tier may be automatically located to a physical storage device of said first storage tier. The criteria may include at least one of capacity limits and performance limits. The first storage tier may include one or more solid state memory devices and each of said one or more other storage tiers nay include one or more physical devices having lower performance characteristics than said one or more solid state memory devices. The one or more other storage tiers may be storage tiers of rotating disk devices. Each of the plurality of scores may be calculated using a plurality of short term and long term metrics. Each of the plurality of scores for one of said plurality of data portions may be calculated using a set of metrics including a first metric indicating an expected rate at which random read miss operations are predicted to occur for said one data portion, a second metric indicating an expected rate at which sequential read misses are predicted to occur for said one data portion, and a third metric indicating an expected rate at which write operations are predicted to occur for said one data portion. When determining the first set of data portions for said first storage tier, said first, second and third metrics may each be associated with an operation type weight indicating a weighted relative preference given to a random read miss operation, a write operation, and a sequential read miss operation based on expected rates at which each operation type is expected to occur. When determining said first set of data portions for said first storage tier and said first storage tier includes solid state drives, said first metric may be associated with a size weight determined in accordance with an expected I/O size for random read miss operations, said second metric may be associated with a size weight determined in accordance with an expected I/O size for sequential read miss operations, and said third metric may be associated with a size weight determined in accordance with an expected I/O size for write operations. The size weights for said first, second and third metrics may be one (1) if said first storage tier is a storage tier of rotating disk drives rather than solid state drives. The plurality of scores may be promotion scores and the method may further include determining a promotion threshold associated with a first position in a ranked list of said plurality of data portions determined by said ranking, each data portion having an associated score above the promotion threshold being moved to said first storage tier if currently included in another storage tier. The method may also include determining a plurality of demotion scores associated with said plurality of data portions; ranking said plurality of data portions in accordance with said plurality of demotion scores; and determining a demotion threshold associated with a position in a second ranked list of said plurality of data portions determined by said ranking in accordance with said plurality of demotion scores, wherein each data portion having a position below said demotion threshold is a candidate for demotion to be moved from said first storage tier to said another storage tier if currently included in said first storage tier. The logical devices may include one or more thin devices, each of said thin devices being a virtually provisioned device, wherein each of said thin devices has logical address range representing a presented storage capacity of said each thin device, and wherein at least a portion of said logical address range is not mapped to physical storage indicating that physical storage is not allocated for said portion. The method may be performed in connection with optimisation processing to optimize data storage system performance. The criteria may include performance limits for each of the plurality of storage tiers. The performance limits may include a response time limit for each of said plurality of storage tiers.
In accordance with another aspect of the invention is a computer readable medium comprising code stored thereon for configuring data storage, the computer readable medium comprising code for: receiving criteria identifying limits for a plurality of storage tiers for one or more logical devices, said plurality of storage tiers including a first storage tier and one or more other storage tiers, said first storage tier being a higher performance storage tier than said one or more other storage tiers; and performing first processing to determine a first set of data portions of the one or more logical devices to be stored on physical devices of the first storage tier subject to limits included in said criteria, wherein said first processing further includes: determining a plurality of scores for a plurality of data portions of the one or more logical devices, wherein one of said plurality of scores is determined for a different one of the plurality of data portions, each of said plurality of scores being calculated using one or more metrics including a first metric weighted based on an expected I/O size; ranking said plurality of data portions in accordance with said plurality of scores; and selecting said first set of data portions from said plurality of data portions based on said ranking. Other data portions of said plurality of data portions not included in the first set may be stored on physical devices of said one or more other storage tiers. Each data portion in the first set may have an associated score that is higher than other scores associated with others of said plurality of data portions of the one or more logical devices not included in the first set.
Features and advantages of the present invention will become more apparent from the following detailed description of exemplary embodiments thereof taken in conjunction with the accompanying drawings in which:
Referring to
Each of the host systems 14a-14n and the data storage system 12 included in the system 10 may be connected to the communication medium 18 by any one of a variety of connections as may be provided and supported in accordance with the type of communication medium 18. The processors included in the host computer systems 14a-14n may be any one of a variety of proprietary or commercially available single or multi-processor system, such as an Intel-based processor, or other type of commercially available processor able to support traffic in accordance with each particular embodiment and application.
It should be noted that the particular examples of the hardware and software that may be included in the data storage system 12 are described herein in more detail, and may vary with each particular embodiment. Each of the host computers 14a-14n and data storage system may all be located at the same physical site, or, alternatively, may also be located in different physical locations. Examples of the communication medium that may be used to provide the different types of connections between the host computer systems and the data storage system of the system 10 may use a variety of different communication protocols such as SCSI, Fibre Channel, iSCSI, and the like. Some or all of the connections by which the hosts and data storage system may be connected to the communication medium may pass through other communication devices, such switching equipment that may exist such as a phone line, a repeater, a multiplexer or even a satellite.
Each of the host computer systems may perform different types of data operations in accordance with different types of tasks. In the embodiment of
It should be noted that although element 12 is illustrated as a single data storage system, such as a single data storage array, element 12 may also represent, for example, multiple data storage arrays alone, or in combination with, other data storage devices, systems, appliances, and/or components having suitable connectivity, such as in a SAN, in an embodiment using the techniques herein. It should also be noted that an embodiment may include data storage arrays or other components from one or more vendors. In subsequent examples illustrated the techniques herein, reference may be made to a single data storage array by a vendor, such as by EMC Corporation of Hopkinton, Mass. However, as will be appreciated by those skilled in the art, the techniques herein are applicable for use with other data storage arrays by other vendors and with other components than as described herein for purposes of example.
The data storage system 12 may be a data storage array including a plurality of data storage devices 16a-16n. The data storage devices 16a-16n may include one or more types of data storage devices such as, for example, one or more disk drives and/or one or more solid state drives (SSDs). An SSD is a data storage device that uses solid-state memory to store persistent data. An SSD using SRAM or DRAM, rather than flash memory, may also be referred to as a RAM drive. SSD may refer to solid state electronics devices as distinguished from electromechanical devices, such as hard drives, having moving parts. Flash devices or flash memory-based SSDs are one type of SSD that contains no moving parts. As described in more detail in following paragraphs, the techniques herein may be used in an embodiment in which one or more of the devices 16a-16n are flash drives or devices. More generally, the techniques herein may also be used with any type of SSD although following paragraphs may make reference to a particular type such as a flash device or flash memory device.
The data storage array may also include different types of adapters or directors, such as an HA 21 (host adapter), RA 40 (remote adapter), and/or device interface 23. Each of the adapters may be implemented using hardware including a processor with local memory with code stored thereon for execution in connection with performing different operations. The HAs may be used to manage communications and data operations between one or more host systems and the global memory (GM). In an embodiment, the HA may be a Fibre Channel Adapter (FA) or other adapter which facilitates host communication. The HA 21 may be characterized as a front end component of the data storage system which receives a request from the host. The data storage array may include one or more RAs that may be used, for example, to facilitate communications between data storage arrays. The data storage array may also include one or more device interfaces 23 for facilitating data transfers to/from the data storage devices 16a-16n. The data storage interfaces 23 may include device interface modules, for example, one or more disk adapters (DAs) (e.g., disk controllers), adapters used to interface with the flash drives, and the like. The DAs may also be characterized as back end components of the data storage system which interface with the physical data storage devices.
One or more internal logical communication paths may exist between the device interfaces 23, the RAs 40, the HAs 21, and the memory 26. An embodiment, for example, may use one or more internal busses and/or communication modules. For example, the global memory portion 25b may be used to facilitate data transfers and other communications between the device interfaces, HAs and/or RAs in a data storage array. In one embodiment, the device interfaces 23 may perform data operations using a cache that may be included in the global memory 25b, for example, when communicating with other device interfaces and other components of the data storage array. The other portion 25a is that portion of memory that may be used in connection with other designations that may vary in accordance with each embodiment.
The particular data storage system as described in this embodiment, or a particular device thereof, such as a disk or particular aspects of a flash device, should not be construed as a limitation. Other types of commercially available data storage systems, as well as processors and hardware controlling access to these particular devices, may also be included in an embodiment.
Host systems provide data and access control information through channels to the storage systems, and the storage systems may also provide data to the host systems also through the channels. The host systems do not address the drives or devices 16a-16n of the storage systems directly, but rather access to data may be provided to one or more host systems from what the host systems view as a plurality of logical devices or logical volumes (LVs). The LVs may or may not correspond to the actual physical devices or drives 16a-16n. For example, one or more LVs may reside on a single physical drive or multiple drives. Data in a single data storage system, such as a single data storage array, may be accessed by multiple hosts allowing the hosts to share the data residing therein. The HAs may be used in connection with communications between a data storage array and a host system. The RAs may be used in facilitating communications between two data storage arrays. The DAs may be one type of device interface used in connection with facilitating data transfers to/from the associated disk drive(s) and LV(s) residing thereon. A flash device interface may be another type of device interface used in connection with facilitating data transfers to/from the associated flash devices and LV(s) residing thereon. It should be noted that an embodiment may use the same or a different device interface for one or more different types of devices than as described herein.
The device interface, such as a DA, performs I/O operations on a drive 16a-16n. In the following description, data residing on an LV may be accessed by the device interface following a data request in connection with I/O operations that other directors originate. Data may be accessed by LV in which a single device interface manages data requests in connection with the different one or more LVs that may reside on a drive 16a-16n. For example, a device interface may be a DA that accomplishes the foregoing by creating job records for the different LVs associated with a particular device. These different job records may be associated with the different LVs in a data structure stored and managed by each device interface.
Also shown in
It should be noted that a service processor 22a may exist external to the data storage system 12 and may communicate with the data storage system 12 using any one of a variety of communication connections. In one embodiment, the service processor 22a may communicate with the data storage system 12 through three different connections, a serial port, a parallel port and using a network interface card, for example, with an Ethernet connection. Using the Ethernet connection, for example, a service processor may communicate directly with DAs and HAs within the data storage system 12.
Referring to
The representation of
With reference back to
As described above, the data storage system 12 may be a data storage array including a plurality of data storage devices 16a-16n in which one or more of the devices 16a-16n are flash memory devices employing one or more different flash memory technologies. In one embodiment, the data storage system 12 may be a Symmetrix® DMX™ or VMAX™ data storage array by EMC Corporation of Hopkinton, Mass. In the foregoing data storage array, the data storage devices 16a-16n may include a combination of disk devices and flash devices in which the flash devices may appear as standard Fibre Channel (FC) drives to the various software tools used in connection with the data storage array. The flash devices may be constructed using nonvolatile semiconductor NAND flash memory. The flash devices may include one or more SLC (single level cell) devices and/or MLC (multi level cell) devices.
It should be noted that the techniques herein may be used in connection with flash devices comprising what may be characterized as enterprise-grade or enterprise-class flash drives (EFDs) with an expected lifetime (e.g., as measured in an amount of actual elapsed time such as a number of years, months, and/or days) based on a number of guaranteed write cycles, or program cycles, and a rate or frequency at which the writes are performed. Thus, a flash device may be expected to have a usage measured in calendar or wall clock elapsed time based on the amount of time it takes to perform the number of guaranteed write cycles. The techniques herein may also be used with other flash devices, more generally referred to as non-enterprise class flash devices, which, when performing writes at a same rate as for enterprise class drives, may have a lower expected lifetime based on a lower number of guaranteed write cycles.
The techniques herein may be generally used in connection with any type of flash device, or more generally, any SSD technology. The flash device may be, for example, a flash device which is a NAND gate flash device, NOR gate flash device, flash device that uses SLC or MLC technology, and the like, as known in the art. In one embodiment, the one or more flash devices may include MLC flash memory devices although an embodiment may utilize MLC, alone or in combination with, other types of flash memory devices or other suitable memory and data storage technologies. More generally, the techniques herein may be used in connection with other SSD technologies although particular flash memory technologies may be described herein for purposes of illustration.
An embodiment in accordance with techniques herein may have one or more defined storage tiers. Each tier may generally include physical storage devices or drives having one or more attributes associated with a definition for that tier. For example, one embodiment may provide a tier definition based on a set of one or more attributes. The attributes may include any one or more of a storage type or storage technology, a type of data protection, device performance characteristic(s), storage capacity, and the like. The storage type or technology may specify whether a physical storage device is an SSD drive (such as a flash drive), a particular type of SSD drive (such using flash or a form of RAM), a type of magnetic disk or other non-SSD drive (such as an FC disk drive, a SATA (Serial Advanced Technology Attachment) drive), and the like. Data protection may specify a type or level of data storage protection such, for example, as a particular RAID level (e.g., RAID1, RAID-5 3+1, RAIDS 7+1, and the like). Performance characteristics may relate to different performance aspects of the physical storage devices of a particular type or technology. For example, there may be multiple types of FC disk drives based on the RPM characteristics of the FC disk drives (e.g., 10K RPM FC drives and 15K RPM FC drives) and FC disk drives having different RPM characteristics may be included in different storage tiers. Storage capacity may specify the amount of data, such as in bytes, that may be stored on the drives. An embodiment may allow a user to define one or more such storage tiers. For example, an embodiment in accordance with techniques herein may define two storage tiers including a first tier of all SSD drives and a second tier of all non-SSD drives. As another example, an embodiment in accordance with techniques herein may define three storage tiers including a first tier of all SSD drives which are flash drives, a second tier of all FC drives, and a third tier of all SATA drives. The foregoing are some examples of tier definitions and other tier definitions may be specified in accordance with techniques herein.
Referring to
Included in the service processor 22a is performance data monitoring software 134 which gathers performance data about the data storage system 12 through the connection 132. The performance data monitoring software 134 gathers and stores performance data and forwards this to the optimizer 138 which further stores the data in the performance data file 136. This performance data 136 may also serve as an input to the optimizer 138 which attempts to enhance the performance of I/O operations, such as those I/O operations associated with data storage devices 16a-16n of the system 12. The optimizer 138 may take into consideration various types of parameters and performance data 136 in an attempt to optimize particular metrics associated with performance of the data storage system 12. The performance data 136 may be used by the optimizer to determine metrics described and used in connection with techniques herein. The optimizer may access the performance data, for example, collected for a plurality of LVs when performing a data storage optimization. The performance data 136 may be used in determining a workload for one or more physical devices, logical devices or volumes (LVs) serving as data devices, thin devices (described in more detail elsewhere herein) or other virtually provisioned devices, portions of thin devices, and the like. The workload may also be a measurement or level of “how busy” a device is, for example, in terms of I/O operations (e.g., I/O throughput such as number of I/Os/second, response time (RT), and the like).
The response time for a storage device or volume may be based on a response time associated with the storage device or volume for a period of time. The response time may based on read and write operations directed to the storage device or volume. Response time represents the amount of time it takes the storage system to complete an I/O request (e.g., a read or write request). Response time may be characterized as including two components: service time and wait time. Service time is the actual amount of time spent servicing or completing an I/O request after receiving the request from a host via an HA 21, or after the storage system 12 generates the I/O request internally. The wait time is the amount of time the I/O request spends waiting in line or queue waiting for service (e.g., prior to executing the I/O operation).
It should be noted that the operations of read and write with respect to an LV, thin device, and the like, may be viewed as read and write requests or commands from the DA 23, controller or other backend physical device interface. Thus, these are operations may also be characterized as a number of operations with respect to the physical storage device (e.g., number of physical device reads, writes, and the like, based on physical device accesses). This is in contrast to observing or counting a number of particular type of I/O requests (e.g., reads or writes) as issued from the host and received by a front end component such as an HA 21. To illustrate, a host read request may not result in a read request or command issued to the DA if there is a cache hit and the requested data is in cache. The host read request results in a read request or command issued to the DA 23 to retrieve data from the physical drive only if there is a read miss. Furthermore, when writing data of a received host I/O request to the physical device, the host write request may result in multiple reads and/or writes by the DA 23 in addition to writing out the host or user data of the request. For example, if the data storage system implements a RAID data protection technique, such as RAID-5, additional reads and writes may be performed such as in connection with writing out additional parity information for the user data. Thus, observed data gathered to determine workload, such as observed numbers of reads and writes, may refer to the read and write requests or commands performed by the DA. Such read and write commands may correspond, respectively, to physical device accesses such as disk reads and writes that may result from a host I/O request received by an HA 21.
The optimizer 138 may perform processing of the techniques herein set forth in following paragraphs to determine how to allocate or partition physical storage in a multi-tiered environment for use by multiple applications. The optimizer 138 may also perform other processing such as, for example, to determine what particular portions of thin devices to store on physical devices of a particular tier, evaluate when to migrate or move data between physical drives of different tiers, and the like. It should be noted that the optimizer 138 may generally represent one or more components that perform processing as described herein as well as one or more other optimizations and other processing that may be performed in an embodiment.
Described in following paragraphs are techniques that may be performed to determine promotion and demotion thresholds (described below in more detail) used in determining what data portions of thin devices to store on physical devices of a particular tier in a multi-tiered storage environment. Such data portions of a thin device may be automatically placed in a storage tier where the techniques herein have determined the storage tier is best to service that data in order to improve data storage system performance. The data portions may also be automatically relocated or migrated to a different storage tier as the work load and observed performance characteristics for the data portions change over time. In accordance with techniques herein, analysis of performance data for data portions of thin devices may be performed in order to determine whether particular data portions should have their data contents stored on physical devices located in a particular storage tier. The techniques herein may take into account how “busy” the data portions are in combination with defined capacity limits and defined performance limits (e.g., such as I/O throughput or I/Os per unit of time, response time, utilization, and the like) associated with a storage tier in order to evaluate which data to store on drives of the storage tier. The foregoing defined capacity limits and performance limits may be used as criteria to determine promotion and demotion thresholds based on projected or modeled I/O workload of a storage tier. Different sets of performance limits, also referred to as comfort performance zones or performance zones, may be evaluated in combination with capacity limits based on one or more overall performance metrics (e.g., average response time across all storage tiers for one or more storage groups) in order to select the promotion and demotion thresholds for the storage tiers.
Promotion may refer to movement of data from a first storage tier to a second storage tier where the second storage tier is characterized as having devices of higher performance than devices of the first storage tier. Demotion may refer generally to movement of data from a first storage tier to a second storage tier where the first storage tier is characterized as having devices of higher performance than devices of the second storage tier. As such, movement of data from a first tier of flash devices to a second tier of FC devices and/or SATA devices may be characterized as a demotion and movement of data from the foregoing second tier to the first tier a promotion. The promotion and demotion thresholds refer to thresholds used in connection with data movement.
As described in following paragraphs, one embodiment may use an allocation policy specifying an upper limit or maximum threshold of storage capacity for each of one or more tiers for use with an application. The partitioning of physical storage of the different storage tiers among the applications may be initially performed using techniques herein in accordance with the foregoing thresholds of the application's allocation policy and other criteria. In accordance with techniques herein, an embodiment may determine amounts of the different storage tiers used to store an application's data, and thus the application's storage group, subject to the allocation policy and other criteria. Such criteria may also include one or more performance metrics indicating a workload of the application. For example, an embodiment may determine one or more performance metrics using collected or observed performance data for a plurality of different logical devices, and/or portions thereof, used by the application. Thus, the partitioning of the different storage tiers among multiple applications may also take into account the workload or how “busy” an application is. Such criteria may also include capacity limits specifying how much of each particular storage tier may be used to store data for the application's logical devices. As described in various embodiments herein, the criteria may include one or more performance metrics in combination with capacity limits, performance metrics alone without capacity limits, or capacity limits alone without performance metrics. Of course, as will be appreciated by those of ordinary skill in the art, such criteria may include any of the foregoing in combination with other suitable criteria.
As an example, the techniques herein may be described with reference to a storage environment having three storage tiers—a first tier of only flash drives in the data storage system, a second tier of only FC disk drives, and a third tier of only SATA disk drives. In terms of performance, the foregoing three tiers may be ranked from highest to lowest as follows: first, second, and then third. The lower the tier ranking, the lower the tier's performance characteristics (e.g., longer latency times, capable of less I/O throughput/second/GB (or other storage unit), and the like). Generally, different types of physical devices or physical drives have different types of characteristics. There are different reasons why one may want to use one storage tier and type of drive over another depending on criteria, goals and the current performance characteristics exhibited in connection with performing I/O operations. For example, flash drives of the first tier may be a best choice or candidate for storing data which may be characterized as I/O intensive or “busy” thereby experiencing a high rate of I/Os to frequently access the physical storage device containing the LV's data. However, flash drives tend to be expensive in terms of storage capacity. SATA drives may be a best choice or candidate for storing data of devices requiring a large storage capacity and which are not I/O intensive with respect to access and retrieval from the physical storage device. The second tier of FC disk drives may be characterized as “in between” flash drives and SATA drives in terms of cost/GB and I/O performance. Thus, in terms of relative performance characteristics, flash drives may be characterized as having higher performance than both FC and SATA disks, and FC disks may be characterized as having a higher performance than SATA.
Since flash drives of the first tier are the best suited for high throughput/sec/GB, processing may be performed to determine which of the devices, and portions thereof, are characterized as most I/O intensive and therefore may be good candidates to have their data stored on flash drives. Similarly, the second most I/O intensive devices, and portions thereof, may be good candidates to store on FC disk drives of the second tier and the least I/O intensive devices may be good candidates to store on SATA drives of the third tier. As such, workload for an application may be determined using some measure of I/O intensity, performance or activity (e.g., I/O throughput/second, percentage of read operation, percentage of write operations, response time, etc.) of each device used for the application's data. Some measure of workload may be used as a factor or criterion in combination with others described herein for determining what data portions are located on the physical storage devices of each of the different storage tiers.
The volumes 110-113 may be provided in multiple storage tiers (TIERS 0-3) that may have different storage characteristics, such as speed, cost, reliability, availability, security and/or other characteristics. As described above, a tier may represent a set of storage resources, such as physical storage devices, residing in a storage platform. Examples of storage disks that may be used as storage resources within a storage array of a tier may include sets SATA disks, FC disks and/or EFDs, among other known types of storage devices.
According to various embodiments, each of the volumes 110-113 may be located in different storage tiers. Tiered storage provides that data may be initially allocated to a particular fast volume/tier, but a portion of the data that has not been used over a period of time (for example, three weeks) may be automatically moved to a slower (and perhaps less expensive) tier. For example, data that is expected to be used frequently, for example database indices, may be initially written directly to fast storage whereas data that is not expected to be accessed frequently, for example backup or archived data, may be initially written to slower storage. In an embodiment, the system described herein may be used in connection with a Fully Automated Storage Tiering (FAST) product produced by EMC Corporation of Hopkinton, Mass., that provides for the optimization of the use of different storage tiers including the ability to easily create and apply tiering policies (e.g., allocation policies, data movement policies including promotion and demotion thresholds, and the like) to transparently automate the control, placement, and movement of data within a storage system based on business needs. The techniques herein may be used to determine amounts or allocations of each storage tier used by each application based on capacity limits in combination with performance limits.
Referring to
As shown in
Referring to
Each of the entries 156-158 of the table 152 correspond to another table that may contain information for one or more logical volumes, such as thin device logical volumes. For example, the entry 157 may correspond to a thin device table 162. The thin device table 162 may include a header 164 that contains overhead information, such as information identifying the corresponding thin device, information concerning the last used data device and/or other information including counter information, such as a counter that keeps track of used group entries (described below). The header information, or portions thereof, may be available globally to the data storage system.
The thin device table 162 may include one or more group elements 166-168, that contain information corresponding to a group of tracks on the data device. A group of tracks may include one or more tracks, the number of which may be configured as appropriate. In an embodiment herein, each group has sixteen tracks, although this number may be configurable.
One of the group elements 166-168 (for example, the group element 166) of the thin device table 162 may identify a particular one of the data devices 61-67 having a track table 172 that contains further information, such as a header 174 having overhead information and a plurality of entries 176-178 corresponding to each of the tracks of the particular one of the data devices 61-67. The information in each of the entries 176-178 may include a pointer (either direct or indirect) to the physical address on one of the physical disk drives of the data storage system that maps to the logical address(es) of the particular one of the data devices 61-67. Thus, the track table 162 may be used in connection with mapping logical addresses of the logical devices corresponding to the tables 152, 162, 172 to physical addresses on the disk drives or other physical devices of the data storage system.
The tables 152, 162, 172 may be stored in the global memory 25b of the data storage system. In addition, the tables corresponding to particular logical devices accessed by a particular host may be stored (cached) in local memory of the corresponding one of the HA's. In addition, an RA and/or the DA's may also use and locally store (cache) portions of the tables 152, 162, 172.
Referring to
Accordingly, a thin device presents a logical storage space to one or more applications running on a host where different portions of the logical storage space may or may not have corresponding physical storage space associated therewith. However, the thin device is not mapped directly to physical storage space. Instead, portions of the thin storage device for which physical storage space exists are mapped to data devices, which are logical devices that map logical storage space of the data device to physical storage space on the disk drives or other physical storage devices. Thus, an access of the logical storage space of the thin device results in either a null pointer (or equivalent) indicating that no corresponding physical storage space has yet been allocated, or results in a reference to a data device which in turn references the underlying physical storage space.
Thin devices and thin provisioning are described in more detail in U.S. patent application Ser. No. 11/726,831, filed Mar. 23, 2007 (U.S. Patent App. Pub. No. 2009/0070541 A1), AUTOMATED INFORMATION LIFE-CYCLE MANAGEMENT WITH THIN PROVISIONING, Yochai, EMS-147US, and U.S. Pat. No. 7,949,637, Issued May 24, 2011, Storage Management for Fine Grained Tiered Storage with Thin Provisioning, to Burke, both of which are incorporated by reference herein.
As discussed elsewhere herein, the data devices 61-67 (and other logical devices) may be associated with physical storage areas (e.g., disk drives, tapes, solid state storage, etc.) having different characteristics. In various embodiments, the physical storage areas may include multiple tiers of storage in which each sub-tier of physical storage areas and/or disk drives may be ordered according to different characteristics and/or classes, such as speed, technology and/or cost. The devices 61-67 may appear to a host coupled to the storage device 24 as a logical volume (logical device) containing a contiguous block of data storage, as discussed herein. Accordingly, each of the devices 61-67 may map to storage areas across multiple physical storage drives. The granularity at which the storage system described herein operates may be smaller than at the file level, for example potentially as small as a single byte, but more practically at the granularity of a single logical block or collection of sequential data blocks. A data block may be of any size including file system or database logical block size, physical block, track or cylinder and/or other size. Multiple data blocks may be substantially the same size or different sizes, such as different size data blocks for different storage volumes or different sized data blocks within a single storage volume.
In accordance with techniques herein, an embodiment may allow for locating all of the data of a single logical portion or entity in a same tier or in multiple different tiers depending on the logical data portion or entity. In an embodiment including thin devices, the techniques herein may be used where different portions of data of a single thin device may be located in different storage tiers. For example, a thin device may include two data portions and a first of these two data portions may be identified as a “hot spot” of high I/O activity (e.g., having a large number of I/O accesses such as reads and/or writes per unit of time) relative to the second of these two portions. As such, an embodiment in accordance with techniques herein may have added flexibility in that the first portion of data of the thin device may be located in a different higher performance storage tier than the second portion. For example, the first portion may be located in a tier comprising flash devices and the second portion may be located in a different tier of FC or SATA drives.
Referring to
Each of 206, 208 and 210 may correspond to a tier definition as described elsewhere herein. Element 206 represents a first storage tier of flash drives having a tier capacity limit C1. Element 208 represents a first storage tier of FC drives having a tier capacity limit C2. Element 210 represents a first storage tier of SATA drives having a tier capacity limit C3. Each of C1, C2 and C3 may represent an available or maximum amount of storage capacity in the storage tier that may be physical available in the system. The AP 204 may be associated with one of more SGs such as SG 202. The AP 204 specifies, for an associated SG 202, a capacity upper limit or maximum threshold for one or more storage tiers. Each such limit may identify an upper bound regarding an amount of storage that may be allocated for use by the associated SG. The AP 204 may be associated with one or more of the storage tiers 206, 208 and 210 that may be defined in a multi-tier storage environment. The AP 204 in this example 200 includes limit 204a identifying a maximum or upper limit of storage for tier1, limit 204b identifying a maximum or upper limit of storage for tier2, and limit 204c identifying a maximum or upper limit of storage for tier3. The SG 202 may be based on an SG definition identifying 202a the logical devices, such as TDs included in the SG.
In connection with techniques herein, the maximum limits 204a, 204b and 204c each represent an upper bound of a storage capacity to which an associated SG is subjected to. The techniques herein may be used to partition less than the amount or capacity represented by such limits. An amount of physical storage of a tier allocated for use by an application is allowed to vary up to the tier limit as defined in the AP 204 in accordance with other criteria associated with the application such as, for example, varying application workload. The optimizer may vary the amount of storage in each tier used by an SG 202, and thus an application, based on workload and possibly other criteria when performing a cost benefit analysis, where such amounts are subject to the limits of the SG's AP and also performance limits described in more detail elsewhere herein. At a second point in time, the workloads and possibly other criteria for the applications may change and the optimizer may reparation the storage capacity used by each application subject to the capacity limits of APs and performance limits.
Referring to
With reference back to
Data used in connection with techniques herein, such as the performance data of
It should be noted that movement of data between tiers from a source tier to a target tier may include determining free or unused storage device locations within the target tier. In the event there is an insufficient amount of free of unused storage in the target tier, processing may also include displacing or relocating other data currently stored on a physical device of the target tier. An embodiment may perform movement of data to and/or from physical storage devices using any suitable technique. Also, any suitable technique may be used to determine a target storage device in the target tier where the data currently stored on the target is relocated or migrated to another physical device in the same or a different tier.
One embodiment in accordance with techniques herein may include multiple storage tiers including a first tier of flash devices and one or more other tiers of non-flash devices having lower performance characteristics than flash devices. The one or more other tiers may include, for example, one or more types of disk devices. The tiers may also include other types of SSDs besides flash devices.
As described above, a thin device (also referred to as a virtual provision device) is a device that represents a certain capacity having an associated address range. Storage may be allocated for thin devices in chunks or data portions of a particular size as needed rather than allocate all storage necessary for the thin device's entire capacity. Therefore, it may be the case that at any point in time, only a small number of portions or chunks of the thin device actually are allocated and consume physical storage on the back end (on physical disks, flash or other physical storage devices). A thin device may be constructed of chunks having a size that may vary with embodiment. For example, in one embodiment, a chunk may correspond to a group of 12 tracks (e.g., 12 tracks*64 Kbytes/track=768 Kbytes/chunk). As also noted with a thin device, the different chunks may reside on different data devices in one or more storage tiers. In one embodiment, as will be described below, a storage tier may consist of one or more storage pools. Each storage pool may include multiple LVs and their associated physical devices. With thin devices, a system in accordance with techniques herein has flexibility to relocate individual chunks as desired to different devices in the same as well as different pools or storage tiers. For example, a system may relocate a chunk from a flash storage pool to a SATA storage pool. In one embodiment using techniques herein, a thin device can be bound to a particular storage pool of a storage tier at a point in time so that any chunks requiring allocation of additional storage, such as may occur when writing data to the thin device, result in allocating storage from this storage pool. Such binding may change over time for a thin device.
A thin device may contain thousands and even hundreds of thousands of such chunks. As such, tracking and managing performance data such as one or more performance statistics for each chunk, across all such chunks, for a storage group of thin devices can be cumbersome and consume an excessive amount of resources. Described in following paragraphs are techniques that may be used in connection with collecting performance data about thin devices where such information may be used to determine which chunks of thin devices are most active relative to others. Such evaluation may be performed in connection with determining promotion/demotion thresholds use in evaluating where to locate and/or move data of the different chunks with respect to the different storage tiers in a multi-storage tier environment. In connection with examples in following paragraphs, details such as having a single storage pool in each storage tier, a single storage group, and the like, are provided for purposes of illustration. Those of ordinary skill in the art will readily appreciate the more general applicability of techniques herein in other embodiments such as, for example, having a storage group include a plurality of storage pools, and the like.
Referring to
It should be noted that although the example 700 illustrates only a single storage pool per storage tier, an embodiment may also have multiple storage pools per tier.
Referring to
In connection with collecting statistics characterizing performance, workload and/or activity for a thin device, one approach may be to collect the information per chunk or, more generally, for the smallest level of granularity associated with allocation and deallocation of storage for a thin device. Such statistics may include, for example, a number of reads/unit of time, #writes/unit of time, a number of prefetches/unit of time, and the like. However, collecting such information at the smallest granularity level does not scale upward as number of chunks grows large such as for a single thin device which can have up to, for example 300,000 chunks.
Therefore, an embodiment in accordance with techniques herein may collect statistics on a grouping of “N” chunks also referred to as an extent, where N represents an integer number of chunks, N>0. N may be, for example, 480 in one embodiment. Each extent may represent a consecutive range or portion of the thin device in terms of thin device locations (e.g., portion of the address space or range of the thin device). Note that the foregoing use of consecutive does not refer to physical storage locations on physical drives but rather refers to consecutive addresses with respect to a range of addresses of the thin device which are then mapped to physical device locations which may or may not be consecutive, may be on the same or different physical drives, and the like. For example, in one embodiment, an extent may be 480 chunks (N=480) having a size of 360 MBs (megabytes).
An extent may be further divided into sub extents, where each sub extent is a collection of M chunks. M may be, for example 10 in one embodiment. In one embodiment, the sub-extent size may correspond to the smallest granularity of data movement. In other words, the sub extent size represents the atomic unit or minimum amount of data that can be operated upon when performing a data movement such as between storage tiers.
Referring to
N=# sub extents/extent*M chunks/sub extent EQUATION 1
An embodiment in accordance with techniques herein may collect statistics for each extent and also other information characterizing activity of each sub extent of a thin device. Statistics for each extent may be characterized as either long term or short term. Short term refers to statistics which may reflect performance, workload, and/or I/O activity of an extent with respect to a relatively short window of time. Thus, short term statistics may reflect recent extent activity for such a short time period. In contrast and relative to short term, long term refers to statistics reflecting performance, workload and/or I/O activity of an extent with respect to a longer period of time. Depending on the evaluation being performed, such as by the optimizer, it may be desirable to place greater weight on short term information than long term, or vice versa. Furthermore, the information maintained per sub extent may be used as needed once particular extents of interest have been identified.
Referring to
As will be described in following paragraphs, the short term rates 320, long term rates 330 and sub extent activity bitmap 306 may be used in connection with a variety of different evaluations such as by the optimizer 138. Generally, the activity level information or data for an extent such as illustrated in
In one embodiment, the short term rates 320 for an extent may include a read miss rate (e.g., random read miss (RRM) rate) 322, a write I/O rate 324 and a prefetch rate 326 for the extent. The long term rates 330 for an extent may include a read miss rate 332 (e.g., number of read misses/unit of time, where a read miss refers to a cache miss for a read), a write I/O rate 334 (e.g., number of writes/unit of time) and a prefetch rate 336 (e.g., number of prefetches/unit of time) for the extent. As known in the art, data may be prefetched from a physical device and placed in cache prior to reference or use with an I/O operation. For example, an embodiment may perform sequential stream I/O recognition processing to determine when consecutive portions of a thin device are being referenced. In this case, data of the sequential stream may be prefetched from the physical device and placed in cache prior to usage in connection with a subsequent I/O operation. In connection with a portion of data at a first point in a sequential stream associated with a current I/O operation, data subsequent to the first point may be prefetched such as when obtaining the portion from a physical device in anticipation of future usage with subsequent I/Os. The short term prefetch rate 326, as well as the long term prefetch rate 336, may also be referred to as denoting a number of sequential reads or sequential read miss operations performed since such prefetching may occur in response to determination that a read operation is performed for data which is not in cache (read miss) and the read operation is for data included in a series of sequentially read data portions as described above. The read miss rates 322 and 332 may represent random read miss (RRM) rates where such read misses (e.g., data requested not currently in cache) are associate with read operations not included in connection with reading data of a sequential stream (e.g., all read misses not used in connection with computing 326 and 336).
Each of the foregoing rates of 320 and 330 may be with respect to any unit of time, such as per second, per hour, and the like. In connection with describing elements 302 and 304 in more detail, what will be described is how an embodiment in accordance with techniques herein may determine the short term rates 320 and long term rates 330 using a decay function and decay coefficients.
In an embodiment in accordance with techniques herein, a decay coefficient may be characterized as a weighting factor given to previous activity information. The higher the coefficient, the greater the weight given to previous activity information for the extent. Thus, the adjusted activity level of an extent at a current time, “An”, may be generally represented as a function of a current observed or actual activity level for the current time, “an”, a decay coefficient, “r”, and previous adjusted activity level for the previous time period or sampling period, “An-1”. In connection with the foregoing, “A” may represent an adjusted activity level, “n” may denote the current time period or sampling period and “n−1” may denote the immediately prior or previous time period or sampling period at which the activity for the extent was determined. In other words, “an” is adjusted to take into account previous activity as represented by “An-1” and “An” represents the resulting adjusted value of “an”. With respect to a statistic or metric such as a number or read misses, “an” and “An” may each represent an integer quantity or number of read misses within a current sampling period, “n”. The foregoing may generally be represented as:
An=an+(r*An-1) EQUATION 2
wherein
an is the actual observed activity metric for the current or “nth” sampling period,
“r” is a decay coefficient,
“An” is the adjusted activity metric for the current or “nth” sampling period, and
“An-1” is the adjusted activity metric from the previous or “n−1” sampling period.
Beginning with an initial time period or sampling period, denoted by i=“0” (zero), the adjusted activity A0 may be initially that which is observed, a0. Subsequent observed or actual activity levels may be adjusted as described above. Generally, “ai” may denote an actual or observed value obtained for an activity metric for a sampling period “i”, where “i” is an integer greater than or equal to 0. “Ai” may similarly denote an adjusted activity metric (or adjusted value for “ai”) for a sampling period “i”, where “i” is an integer greater than or equal to 0. Thus, for consecutive sample periods at which actual or observed activity metrics are obtained (as denoted by lower case “ai”s), corresponding adjusted activity levels (e.g., “A” values) may be determined as follows:
A0=a0/*Adjusted activity level A0, at time=0 or initially*/
A1=a1+(r*A0)/*Adjusted activity level A1, at first sampling period, i=1*/
A2=a2+(r*A1)/*Adjusted activity level A2, at second sampling period, i=2*/
:
and so on for subsequent sampling periods 3, 4, and the like, based on EQUATION 2.
In connection with EQUATION 2, 0<=r<1, where “r” is a decay coefficient or weight given to previous activity. Varying “r” in EQUATION 2 results in accordingly varying the weight given to past or previous activity. If r=0, then no weight is given to previous or historic values. Thus, the closer “r” is to 0, the lesser weight given to previous activity. Similarly, the closer “r” is to 1, the greater the weight given to previous activity. In connection with determining an adjusted activity level, An, using EQUATION 2 for short term and long term, different decay coefficients may be selected. Generally “r” for short term is less than “r” used in connection with long term activity. For example, in one embodiment, “r” used in connection short term activity levels may be 50% or 0.50 or smaller. “r” used in connection with long term activity levels may be 80% or 0.80 or larger. The foregoing are exemplary values that may be selected for “r” in connection with short term and long term activity levels depending on the weight to be given to previous activity. In connection with short term activity, a decay coefficient may be selected in accordance with providing a relatively short term rate of decay for an activity level metric determined at a point in time. For example, a short term rate of decay may provide for a rate of decay for an activity level metric on the order of one or more hours (e.g., less than a day). In connection with long term activity, a decay coefficient may be selected in accordance with providing a relatively long term rate of decay for an activity level metric determined at a point in time. For example, a long term rate of decay may provide for a rate of decay for an activity level metric on the order of one or more days, a week, and the like. Thus, an activity metric at a first point in time may have a weighted or residual effect on an adjusted activity level determined at a later point in time in accordance with the selected decay coefficient indicating the rate of decay of the activity metric.
As mentioned above, EQUATION 2 results in a metric or count, such as a number of read misses, number of writes, or number or prefetches during a sample period. It may be desirable to also determine a rate with respect to a unit of time, such as per second, per hour, and the like, for each of the foregoing adjusted activity metrics An. A rate with respect to a unit of time for the adjusted activity level An may be represented as:
Ar=An*(1−r)/(1−rn-1) EQUATION 3
where Ar=the adjusted activity rate per unit of time,
r=decay coefficient or weight as described above,
n=denotes an “nth” sampling period as described above,
An=adjusted activity level determined for a given sampling period “n” (e.g. using EQUATION 2 as described above).
Generally, the higher the decay coefficient, r, the slower the change in Ar as may be the desired case with long term Ar values. Thus an embodiment may select decay coefficients for use with long term and short term Ar values so that, when plotted with respect to time, long term Ar values generally have a smaller slope than that associated with short term Ar values.
Referring to
In one embodiment, each of the different An values determined using EQUATION 2 may be converted to a corresponding Ar value using EQUATION 3 when desired.
In connection with the foregoing, for example, with respect to a number of read misses, “an” represents the number of such operations that have occurred in a current sample period, n. For example, if a sample period=10 minutes so that statistics for an extent are collected and/or computed every 10 minutes, “an” represents the number of read misses that occurred in the last 10 minute sample period or time interval. An-1 represents the previous or last A calculation (e.g., as determined using EQUATION 2) from the previous sample period, denoted “n−1”.
With reference back to
The short term information 302 may also include storing previous A values as determined for the sampling period “n−1” using EQUATION 2 above. For example, short term information 302 may also include storing three (3) previous adjusted activity level values or A values for sampling period “n−1” using EQUATION 2 above for the read miss count, prefetch count and write count.
The short term information 302 may also include a timestamp value indicating the timestamp associated with the previous sampling period “n−1”.
Using the above-mentioned short term information 302, an embodiment may calculate updated short term rates 320 using EQUATION 3 for a sampling period “n” for a selected “r” as a short term decay coefficient. With each new sampling period, the short term information may be accordingly updated so that which is associated with sampling period “n” subsequently becomes associated with sampling period “n−1”.
The long term information 304 may include long term rates or Ar values as determined using EQUATION 3 for a read miss rate (e.g., number of read misses/second), a prefetch rate (e.g., number of prefetches/second) and a write rate (e.g., number of writes/second). The long term information 304 may also include a time duration interval used for determining an adjusted Ar value for the current time or sampling period “n”. For example, the time duration interval may represent the amount of time for which statistics are collected and used in connection with long term Ar values. An embodiment may store a set of long term Ar values rather than calculate such Ar values on demand from other stored information as in the case above for short term rates 320 (e.g., where short term information 302 is stored and used to calculate short term rates 320 on demand). Thus, in such an embodiment, the long term rates 330 may be included the long term information 304 where such long term rates 330 may be updated with each sampling period. In one embodiment with the arrival of a new sampling period “n”, the long term information 304 may include Ar values for the foregoing statistics as determined using EQUATION 3 for a sampling period “n−1”. These long term Ar values for “n−1” may each be multiplied by the time duration interval to determine An-1, an adjusted metric for the long term time period. The foregoing An-1 value may then be used with EQUATION 2 to determine An for the current sampling period “n” using a selected “r” as a long term decay coefficient. Using An, EQUATION 3 may then be used to obtain updated long term rates Ar values. With each new sampling period, the long term information may be accordingly updated so that which is associated with sampling period “n” subsequently becomes associated with sampling period “n−1”.
With reference back to
Referring to
The activity bitmap may be used in connection with determining an activity level associated with each sub extent, the smallest amount of data that can be associated with a data movement operation to relocate data from one physical device to another. It should be noted that an embodiment may have functionality and capability to physically move data in units or amounts less than a sub extent. However, when performing processing to determine data movement candidates, such as by the optimizer, such processing may consider candidates for data movement which have a minimum size of a sub extent. That is, all data of the sub extent may be either moved or relocated as a complete unit, or remains in its current location. In connection with a sub extent when performing a data movement, it may be that not all chunks of the sub extent are actually moved. For example, suppose a sub extent is 10 chunks and the sub extent is to be moved from a first storage tier, such as from SATA or FC, to a second storage tier, such as flash. It may be that 9/10 chunks of the sub extent are unallocated or already in flash storage with only 1 chunk stored in the first storage tier. In this case, processing only needs to actually move the single chunk from the first storage tier to flash since the remaining 9 chunks are either already in the flash tier or unallocated. With a sub extent, the amount of data actually moved may be at most the size of the sub extent but may be less depending on, for example, whether all chunks of the thin device sub extent are allocated (e.g., actually map to physical storage), depending on the current physical device upon which chunks of the sub extent are located prior to movement, and the like. It should be noted that chunks of a sub extent may be located in different storage tiers, for example, depending on where the chunk's data is stored such as at the time when written as well as the result of other conditions that may vary with embodiment.
As an example use of the activity bitmap is in connection with promotion and demotion. As an example use of the activity bitmap, the bitmap may be used to determine selective sub extents which exhibit the highest activity level such as those having counters=3 (e.g., “hot” or active areas of the extent). These sub extents may be candidates for promotion or data movement to a higher performing storage tier and may be given preference for such promotion and data movement over other sub extents having activity bitmap entries which are less than 3. In a similar manner, the activity bitmap may be used to identify the “coldest” or inactive sub extents. For example, sub extents having bit map entries=0 may be candidates for demotion to a lower performing storage tier.
In connection with promotion data movements, an embodiment may want to be responsive to a change in workload with respect to the short term. With demotion, an embodiment may not want to move data as quickly as with promotion and may also want to consider longer term workloads prior to moving such data to a lesser performing storage tier. With promotion, an embodiment may give greater weight to ST workload and activity data. With demotion, an embodiment may additionally consider LT workload and activity rather than just such ST information.
The information as described and illustrated in
As another example, the activity bitmaps of extents may be used to determine a promotion ranking used to identify which extent may be promoted prior to one or more other extents. To further illustrate, an embodiment may have two extents, both which are candidates for promotion. The two extents may exhibit similar activity levels based on aggregate extent-level information such as based on short term rates 320 for each extent. The extent having the lesser number of active sub extents may have a higher priority for movement than the other extent. For example, processing may be performed to count the number of non-zero bit map entries for each of the two extents. The extent having the lower count may have a higher priority than the other extent having a higher count. In other words, the extents may be ranked or ordered for promotion based on a number or count of non-zero bit map entries. The extent having the lower count may be characterized as also exhibiting the greatest activity level density based on the foregoing counts of the activity bitmaps.
As another example in connection with demotion, an embodiment may use one or more of the short term rates 320 in combination with one or more of the long term rates 330 to identify one or more inactive extents based on such aggregated extent-level activity data. Subsequently, once an inactive extent is identified, the extent's activity bitmap may be examined to determine which sub extents are inactive and should be demoted rather than automatically demoting all sub extents of the inactive extent. Processing may be performed to selectively move some of the sub extents (e.g., those with counters=0, counters less than some threshold such as 1, and the like) to a lower performing storage tier.
One embodiment in accordance with techniques herein may include multiple storage tiers including a first tier of flash devices and one or more other tiers of non-flash devices having lower performance characteristics than flash devices. The one or more other tiers may include, for example, one or more types of disk devices. The tiers may also include other types of SSDs besides flash devices.
The different levels of activity information described herein as obtained at a thin device level, extent level, and sub extent level provide a hierarchical view for characterizing activity of different portions of thin devices. Activity information at higher device levels may be used to first identify devices which may be candidates for data movement, such as between storage tiers (e.g. for promotion and/or demotion). In connection with thin devices, once such a first device is identified, additional detail regarding the first device's activity as reflected in extent activity level information may be used to identify an extent of the first device as a candidate for data movement. Subsequently, the activity bitmap for the extent identified may then be used to determine one or more sub extents of the identified extent for data movement. The techniques herein may be used for collecting and tracking activity of thin devices. Use of the decay coefficients and equations for determining adjusted activity levels to account for previous activity levels provides an effective way of tracking workload and activity over time without having to keep a large database of historical statistics and metrics for long and short time periods.
In addition to the activity information described above for each extent and sub extent of a thin device, an embodiment may also track device level activity information for logical devices (e.g., thin devices, LVs, and the like) and physical devices in a data storage system as also noted. Additionally, an embodiment may track activity information for thin device pools. When a DA or other device interface services an I/O, the DA may not typically have any knowledge regarding thin devices as may be known from the host's point of view. In connection with collecting data for use with techniques herein, each DA may be provided with additional mapping information regarding thin devices and where storage for the thin devices is allocated (e.g., such as described by the allocation map). The DA may use this information to determine what thin device (if any) is associated with a given back end I/O request. When the DA is servicing a back end I/O request, the DA may record information about the I/O including information about the thin device associated with the I/O request. Such additional information about the thin device may be used in order to perform statistics collection of activity data for the thin devices in accordance with techniques herein.
In addition to the statistics and activity data described above, an embodiment may also collect and store information regarding expected I/O size information for each extent, thin device (or other logical device), physical device, and the like. Such information may be determined in any one or more suitable ways in an embodiment. For example, an embodiment may determine expected I/O sizes that represent the average size with respect each of the particular types of I/O operations for which statistics are collected. In connection with the embodiment herein, the types of I/O operations for which statistics are collected may be as described above for read miss or random read miss (RRM), prefetch (P) or sequential read miss (SRM), and write (W). In a manner similar to that as described elsewhere herein for other statistics, the average I/O sizes may be determined based on size information collected for observed I/O operations. The collected size information based on observed I/Os may be used in determining or modeling expected I/O sizes in connection with equations, such as EQUATION 4, described elsewhere herein when determining various scores. For example, an embodiment may determine a first average I/O size based on I/O sizes of write operations observed for a time period, a second average I/O size based on I/O sizes for SRM operations for a time period, and a third average I/O size based on I/O sizes for RRM operations for a time period. The foregoing average I/O sizes may be tracked with respect to each extent and other levels (e.g., thin device, physical device, etc) in a manner similar to that as described above such as in
Ave size for given I/O type=TOTAL_DATA_TRANSFER/TOTAL_OPS EQUATION 3A
where
“Ave size for given I/O type” is the average or expected I/O size for a given I/O operation type (e.g., Read, Write, Read miss, etc.);
“TOTAL_DATA_TRANSFER” is the total amount of data (e.g., in bytes, megabytes or other size unit) for the desired I/O operation type for a given time period; and
“TOTAL_OPS” is the total number of I/O operations observed during the time period for which the TOTAL_DATA_TRANSFER is observed.
It should be noted that EQUATION 3A is one way in which an embodiment may estimate that averages as may be used in connection with expected I/O sizes as described elsewhere herein. Another way an embodiment may determined average I/O sizes is based on a an equation using weighted averages, using information as may be gathered using the allocation map as described elsewhere herein (e.g., to gather information for data portions based on I/Os directed to the physical device where such data portions are stored), and more generally any suitable technique.
In connection with techniques in following paragraphs, the extent-based short term and long term statistics or metrics as described in
((P1*P7*s_rrm)+(P2*P8*s_w)+(P3*P9*s_p)+(P4*P10*l_rrm)+(P5*P11*l_w)+(P6*P12*l_p))/(# Active Subext+1)
where s_rrm is the rate of short term random read misses (322), s_w is the rate of short term writes (324), s_p is the rate of short term pre-fetches or SRMs (326), l_rrm is the rate of long term random read misses (332), l_w is the rate of long term writes (334), and l_p is the rate of long term pre-fetches or SRMs. The coefficients P1-P12 may be set as appropriate and are described in more detail elsewhere herein. It should be noted that “#Active Subext” represents the number of active subextents or subportions of an extent or other larger data portion for which the score is being determined. Examples of evaluating when a subextent or other subportion is active are described elsewhere herein. It should be noted that metrics used in connection with determining promotion and/or demotion score may take into account I/O size.
The coefficients P1-P6 may represent weights assigned to the different operation types denoting how much weight is given to each particular operation type (e.g., which of random read miss (RRM), prefetch (P) or sequential read miss (SRM), write (W)) and the long term and short term variants of expected rates at which such operation types are expected (e.g., predicted or modeled) to occur in the future. In one aspect, the coefficients P1 and P4 represent the weighted preference given to the RRM operation type relative to the other operations types of SRM (or P) and W. In particular, P1 represents the weighted preference given to the short term operation count or rate for the RRM operation type and P4 represents the weighted preference given to the long term operation count or rate for the RRM operation type. In a similar manner, the coefficients P2 and P5 represent the weighted preference given to the W operation type relative to the other operations types of SRM (or P) and RRM. In particular, P2 represents the weighted preference given to the short term operation count or rate for the W operation type and P5 represents the weighted preference given to the long term operation count or rate for the W operation type. Furthermore, the coefficients P3 and P6 represent the weighted preference given to the P or SRM operation type relative to the other operations types of W and RRM. In particular, P3 represents the weighted preference given to the short term operation count or rate for the P or SRM operation type and P6 represents the weighted preference given to the long term operation count or rate for the P or SRM operation type. The weights or coefficients P1-P6 may be generally referred to as operation type weights. In some embodiments, values for P1-P6 may be dynamically selected each time a new set of statistics or metrics (e.g., 320 and 330 of
The coefficients P7-P12 may represent additional weights assigned or associated with each particular variant combination of operation type (e.g., RRM, SRM or P, and W) and short term or long term for each operation type. Generally, the coefficients P7-P12 may represent weighting factors that may be characterized as varying with, dependent on, or a function of, expected I/O size for the three different operation types of RRM, SRM or P, and W. In particular, P7 and P10 represent weighting factors that vary with, or are a function of, expected I/O size for the RRM operation type. P8 and P11 represent weighting factors that vary with, or are a function of, expected I/O size for the W operation type. P9 and P12 represent weighting factors that vary with, or are a function of, expected I/O size for the P or SRM operation type. Weights P7-P12 may also be referred to herein as I/O size or data transfer weights. As noted above, EQUATION 3A is one way in which the expected I/O size may be determined for use in connection with determining such weights. It should also be noted that as a variation to the above where in one embodiment, size weights as represented using P7-P12 may be applied to only the short term metrics (e.g., always use a size weight of 1 for weights P10, P11 and P12 for long term metrics).
In some embodiments, values for P7-P12 may be dynamically selected each time a new set of statistics or metrics (e.g., 320 and 330 of
Values of P7-P12 may be selected as a function of expected I/O sizes for the different operation types. For example, P7 and P10 may be determined as a function of the expected I/O size of the RRM operations with respect to the extents for which the promotion score is being determined. P8 and P11 may be determined as a function of the expected I/O size of the W operations with respect to the extents for which the promotion score is being determined. P9 and P12 may be determined as a function of the expected I/O size of the SRM or P operations with respect to the extents for which the promotion score is being determined.
Values for P7-P12 may be based on each storage tier and drive technology depending on the sensitivity of response time to I/O size. Thus, the drive technology may be used in selecting that particular values for P7-P12 such as using curves of
The demotion score for an extent may be represented in EQUATION 5 as:
(P4*P10*s_rrm)+(P5*P11*s_w)+(P6*P12*s_p)+(P1*P7*l_rrm)+(P2*P8*l_w)+(P3*P9*l_p)
where s_rrm, s_w, p1, etc. are as set forth above.
As noted above in connection with the exemplary EQUATIONS 4 and 5 for computing, respectively, the promotion and demotion scores, the same set of coefficients may be used. Alternatively, an embodiment may, however, use a different set of coefficients for computing the promotion and demotion scores.
In a multi-tiered storage system as described herein, an application having its data stored on thin devices of a storage group may be allowed to use multiple tiers of storage. In order to be able to use the storage of the tiers efficiently and also move a minimal number of chunks between tiers, chunks which are the most active or “hot” need to be located in the higher tiers (e.g., promoted to such tiers if not already located there) and chunks which are least active or “cold” need to be located in lower storage tiers (e.g., demoted to such tiers if not already located there). After identifying the hot and cold chunks, processing may be performed to determine how much of the hot chunks should be placed in the different storage tiers in order to efficiently utilize the higher performing tiers, such as flash tiers, while also avoiding overloading any given tier with I/O request or I/O transfer activity to the point that overall performance (e.g., across all tiers in the AP, across one or more SGs, for the whole data storage system, and the like with respect to the physical devices under consideration) would have been better had less of the workload been placed in the tier. In connection with the foregoing, techniques are described in following paragraphs which determine promotion and demotion thresholds of a data movement policy that may be associated with one or more SGs. The data movement policy as described herein in the context of thin devices affects what data portions of thin devices are data movement candidates and may be moved to another tier. The selection of promotion and demotion thresholds may be made by considering criteria including performance limits (e.g., response time, number of I/Os per time period, and the like) and capacity limits. The performance limits may be flexible or adaptable and specified for each storage tier. The capacity limits may also be specified for each storage tier and may include capacity limits included in an AP for the affected one or more SGs. The techniques model response time of target storage tiers when evaluating different alternative hypothetical considerations in which performance limits are varied for each tier when selecting promotion and demotion thresholds. The different sets of performance limits in combination with capacity limits are evaluated by modeling the expected target tier performance and then determining an overall performance metric representing an aggregate modeled performance metric across all target storage tiers for all affected SGs. In one embodiment, the overall performance metric may be an average response time determined with respect to all target storage tiers using the modeled response time as determined for each such tier. The average response time is used to compare the overall modeled performance for the storage tiers when evaluating different sets of performance limits for each target tier. Each set of performance limits specified for multiple tiers may be used as a way to provide weighting factors for I/O workload distribution across the tiers in order to reflect the performance differences of the different tier storage technologies. Utilizing such “what if” analysis to evaluate different sets of performance limits coupled with capacity limits provides for determining promotion and demotion thresholds that may be used by the DA, or more generally, other backend data storage system components, in connection with performing data movements in accordance with workload or performance impact across all target storage tiers to increase overall performance.
In connection with techniques herein as mentioned above, response time may be considered as performance criteria alone, or in combination with other performance criteria in combination with capacity limits, when determining promotion and demotion thresholds affected what data portions of a thin device may be moved between physical storage devices in different storage tiers. The techniques herein consider different performance characteristic information and curves that may vary with each storage tier, type of physical device, device vendor, and the like. In particular, performance curves for the different storage tiers may be determined and used to model target tier and also overall SG performance across storage tiers as part of processing to evaluate different sets of performance limits in combination with capacity limits. As an example, consider a workload of N I/O operations/second. The response time experienced for the same workload varies with storage tier due to the underlying capabilities of each tier's technology. As such, performance curves may be used in connection with techniques herein to model expected response times if a particular data movement is performed in accordance with candidate promotion and demotion thresholds.
Referring to
In connection with estimating thin device workloads, various metrics that may be used are described herein and also in U.S. patent application Ser. No. 12/924,396, filed Sep. 25, 2010, TECHNIQUES FOR STATISTICS COLLECTION IN CONNECTION WITH DATA STORAGE PERFORMANCE, which is incorporated by reference herein. Workload for thin devices may be determined in a variety of different ways in connection with determining the contributions of the thin device data portions that may be stored in multiple thin device pools. One approach may be to examine the allocation map and determine the workload of data portions based on I/Os directed to the physical device where such data portions are stored. However, an embodiment may use alternative approaches to estimate thin device workload due to additional resources consumed in connection with use of the allocation map which may adversely impact performance. When data portions of a thin device are moved from a first storage tier to a second storage tier, the related workload of such data portions are moved to the target tier. In one embodiment, storage for thin devices may be evenly distributed across a pool of data devices comprising a thin device pool. This results in even distribution of capacity and I/O workload thereby making it possible to correlate I/O workload and capacity allocation at the pool level rather than reading the allocation map for each thin device. In other words, a workload for a thin device data portion having storage allocated from a thin device pool of data devices may be estimated by collecting thin device pool statistics and then apportioning an amount of the workload indicated by the collected data distributed evenly across all data portions stored in the pool.
In connection with
Referring to
Referring to
Referring to
Referring to
Referring to
As such, based on the performance curves of
In connection with estimating modeled response times, performance curves such as that of
The determination of the optimal tier for each extent of storage is driven by the goal of maximizing the chances of achieving storage performance objectives. This goal will tend to be achieved if the storage system can arrange for the largest share possible of the storage request workload to be serviced by the highest performing storage tiers. In connection with techniques herein, scores, such as the promotion and demotion scores, may be metrics or measurements used to drive this determination process where such scores may be calculated for each storage extent. The input to the calculation is information about the expected storage request workload that the storage extent will receive. The promotion and demotion scores as described herein provide a measure of how ‘well suited’ a storage extent is for placement on a first higher performing storage tier as opposed to a second lower performing storage tier. If the promotion and demotion scores used for the storage tiers are defined properly, then for a first promotion score used when determining what extents to store in an EFD storage tier, if storage extent or portion A has a higher promotion score than storage extent B, then storage extent A is better suited for placement on the EFD tier than storage extent B. Of key importance here is how an embodiment quantifies ‘well suited’. Informally, a storage extent should be considered more ‘well-suited’ for a high tier, such as an EFD tier, if placing the storage extent on the high tier tends to allow a greater number of storage requests (especially RRM requests) to be packed into the high tier. The foregoing provides for use of scores or metrics which guide the selection of the best extents to place on the different tiers.
The use of promotion and demotion scores with properly selected weights or coefficients P1-12 allows the storage system to identify the best extents to place in the different storage tiers in a computationally efficient manner. As described herein, for example, processing may be performed to sort the storage extents according to their promotion scores as may be used when determining which extents are best suited for the EFD storage tier. A sufficient number of the extents with the highest promotion scores may be selected to fill the EFD tier based on any one or more of storage capacity limits and/or performance limits for the EFD tier. Similarly, other tiers may be filled with selected extents using scores determined using other values for weights P1-P12 selected for non-EFD tiers. This approach has the important property that it scales well to configurations involving very large numbers of storage extents.
As described above, the promotion and demotion scores may have values selected for P7-P12 (e.g., for the size or data transfer weights) to express any existing dependency of the score on I/O size. Weights P7-P12 may have values selected which are always 1 for non-EFD tiers or, more generally, for those storage tiers having drives which do not have a substantial affect on response time or other performance metric used to measure system performance.
To illustrate how values for P7-P12 may be selected for an EFD tier where such values for the EFD tier may not always be 1 and may depend on the expected storage tier workload characteristics and the particular performance characteristics of the drives in the high tier, consider the case of a multi-tier storage configuration that includes a storage tier comprised of Enterprise Flash Drives (EFDs). Generally speaking, EFD drives are considered high performing because they are particularly efficient at I/O operation processing (much more so than rotating drives are). However, the data transfer rates supported on EFD drives are not correspondingly large in comparison to data transfer rates for rotating disk drives. As known in the art, data transfer rate or throughput may be characterized as the speed at which data can be transferred between devices. For example, data transfer rates may be expressed in terms of Mbps (amount of data transferred for given unit of time). As such, an embodiment may perform processing to select values for P7-P12 dynamically for the EFD tier by considering the expected workload on the EFD storage tier and the specific performance characteristics of the EFD drives underlying the EFD storage tier. For example, if the expected workload on the EFD storage tier is light enough that the data transfer limits of the EFD drives will not be approached, then an embodiment select size weights (e.g., values of P7-P12) for use with EQUATION 4 that are one to thereby allow the greatest number of storage requests to be packed into the available storage capacity in the EFD storage tier (e.g., an embodiment may use a function for EQUATION 4 that assigns a value of 1 to parameters P7-P12). If the expected workload on the EFD storage tier is heavy enough to approach the data transfer limits of the EFD drives, then an embodiment may select size weights (e.g., values of P7-P12) for use with EQUATION 4 that place greater weight on storage requests with small I/O sizes will allow the greatest number of storage requests to be packed into the available data transfer capacity of the EFD storage tier (e.g., an embodiment may use a function for EQUATION 4 that assigns larger values to parameters P7-P12 for data portions with smaller I/O size).
An embodiment in accordance with techniques herein may use models of drive performance to determine whether the performance of the drives in a storage tier is data transfer limited, and the values for P7-P12 may be assigned accordingly as a function of varying I/O size for the different operation types as described elsewhere herein. With scores for the EFD storage tier, there is a preference to have extents in this EFD tier which have higher IOPS of smaller sizes. Therefore, an embodiment may select values for P7-P12 providing a bias or greater weight to a short term or long term metric when the operation type (e.g., RRM, SRM, or W) has smaller expected I/O sizes.
To illustrate how an embodiment may select values for P1-P6 (e.g., operation type weights) for use in the scores for a tier where such values for P1-P6 may depend on the expected storage tier workload characteristics and the particular performance characteristics of the high tier drives, let us again consider the case of a multi-tier storage configuration that includes a storage tier comprised of EFDs. Suppose performance of a particular type of EFD drive is particularly sensitive to the amount of write activity on the drive, with performance degrading for larger amounts of write activity. An embodiment in accordance with techniques herein may take this into account when selecting values for P1-P6. For example, consider a first case where, if the total amount of write workload that may be delivered to the EFD storage tier is light enough to not degrade the performance of the EFD drives, then the scores calculated for the EFD tier may use a small positive (or zero) values for P2 and P5 associated, respectively, with the short term and long term W statistics (e.g., s_w and l_w from EQUATIONS 4 and 5), and relatively large values for P1 and P4 associated, respectively, with the short term and long term RRM statistics (e.g., s_rrm and l_rrm from EQUATIONS 4 and 5), since this allows the greatest number of RRM requests to be packed into available EFD storage. However, consider a second alternative case wherein, if the total amount of write workload that may be delivered to the EFD storage tier is heavy enough to degrade the performance of the EFD drives, then the scores may select lower (e.g., in comparison to those for the first case) or negative values for P2 and P5 associated with the short term and long term W statistics, and yet larger values for P1 and P4 (e.g., larger than in the first case) associated, respectively, with the short and long term RRM statistics. This selection in the second case allows the greatest number of RRM requests to be packed into available EFD storage. An embodiment in accordance with techniques herein may use models of drive performance to determine whether the performance of the drives in a storage tier is write performance limited, and values for P1-P6 may be assigned accordingly.
More generally in connection with selecting values for P1-P6 of the scores, values may be selected depending on how much preference is given to provide better performance (such as better response time) for a particular operation type (e.g., R vs. W, or preference ordering of multiple types RRM, SR, and W). For example, an embodiment may give higher weight to RRM over W since it is more likely that there is an application waiting for the data of the RRM operation to complete before the application can further proceed with processing. An embodiment may give less weight to sequential read (SR) operations than RRM operations. Write operations and associated statistics in the scores may be given the least relative weight because a waiting host or application performing the write may receive an acknowledge that the operation is complete once the write data is written to cache rather than having to wait for data to be read from a physical drive as with any type of read miss operation (e.g., RRM and SR collectively).
As discussed elsewhere herein, policies may be used to determine when to promote data (map the data to a relatively faster tier) and when to demote data (map the data to a relatively slower tier). In particular, one such policy is a data movement policy based on promotion and demotion thresholds that may be determined using promotion and demotion scores for data portions. In an embodiment herein, this may be performed by first determining a score for different portions of a storage space based on relative activity level and then constructing promotion and demotion histograms based on the different scores and the frequency of each. In connection with thin devices, each of the data portions may correspond to a logical extent for which such scores are determined. Exemplary ways in which the promotion and demotion scores may be calculated are described above. The promotion and demotion scores may be used, respectively, in connection with the promotion and demotion histograms described below in more detail. Generally, the scores may be characterized as reflecting the I/O benefit to the host application and cost (e.g., in terms of performance bandwidth) to the targeted storage device tier. In connection with constructing the histogram, all extents are ordered or sorted according to their scores, from highest to lowest. Those extents having the highest scores are generally those preferred to be selected for having storage allocated from the highest performing tier. The histogram is one way in which such scores may be sorted and utilized in connection with techniques herein. It will be appreciated by those of ordinary skill in the art that there are alternative ways to define and compute the scores than as described herein. In one embodiment described herein, the scores may be computed differently for promotion and demotion to reflect the difference in criteria related to data movement into and out of storage tiers.
For purposes of illustration, consider an example of a single SG which may use a group of data devices, and thus physical devices, in three thin device pools—one for each of three storage tiers such as illustrated in
In connection with techniques herein, assume a first set of promotion scores are determined using a first promotion score having weights or coefficients selected for the first or highest storage tier to be filled, the EFD storage tier. A first promotion histogram described below in connection with
Referring to
In a similar manner, for those extents or data portions which do not have first promotion scores at or above S1, a second set of new promotion scores may be determined whereby each promotion score of the second set is based on a second promotion score using values for P1-P12 selected for the next fastest or second storage tier to be filled (e.g., the FC rotating disk drive tier in this example). As described herein, values for P7-P12 may all be 1 with values for P1-P6 selected as may be determined dynamically or based on fixed values. Based on these new second promotion scores, a second promotion histogram is constructed as described in connection with
It should be noted that above-noted two promotion histograms of
It should be noted that an embodiment using a histogram may select a suitable number of bins or buckets and an interval for each such bin. In one embodiment, the size of each bin may be driven by a selected number of bins with each bin having the same size. Additionally, an embodiment may use different techniques in connection with mapping or converting the promotion and demotion scores to indices associated with histogram bins. For example, an embodiment may use linear scaling to set a lower boundary for buckets having an associated index lower than a selected pivot value and may use logarithmic scaling to set a lower boundary for buckets above the pivot. Logarithmic scaling may be appropriate in embodiments having larger scores or a wide range of scores in order to scale the size of scores above the pivot. In such embodiments, the score range associated with a bucket interval above the pivot varies so that a reasonable number of data portions are mapped to the associated bucket. Whether a histogram or other suitable technique is used may vary with the number of buckets, the number of data portions, and the like.
Additionally, it should be noted that rather than have a histogram with frequency on the Y-axis as in
In connection with determining the first tier promotion threshold S1 of
Of course, an embodiment may provide any number of comfort zones more or less than seven and for a different number of storage tiers. Additionally, the foregoing values are exemplary and may vary with technology, drive vendor, and the like. Generally, values specified as the performance limit metrics, such as response times, may vary with the workload and/or other workload characteristics (such as I/O size) of a particular system and may be determined in any suitable manner. For example, values for the foregoing metrics may be made based on knowledge regarding particular workload of a system and typical performance of drives of different storage tiers in a system. In this manner, limits specified may be realistic and in accordance with typical workload performance within a system. It should be noted that the foregoing limits may also be selected based on end user performance requirements. Additionally, as noted elsewhere herein, although response time is used as the workload or performance metric in connection with the foregoing comfort zones, other performance criteria metrics may be used in combination with, or as an alternative to, response time. For example, an embodiment may use utilization as a metric in a manner similar to response time in connection with techniques herein. That is, just as comfort zones include response time limits for storage tiers, comfort zones may include other criteria such as a utilization for each storage tier. As known in the art, utilization of a resource, such as a physical drive or with respect to physical drives of a storage tier, may be defined as a metric measuring an amount of time a device is utilized or in a non-idle state. For example, utilization for a storage tier may be represented as a percentage (e.g., based on a ratio of an amount of time the physical devices of the storage tier are in the non-idle state/total amount of time). The foregoing utilization metric may represent the average utilization for a storage tier determined over a period of time.
Generally, processing may be performed to determine a set of promotion thresholds for the different storage tiers (e.g., S1 of
Similarly, in connection with the above-mentioned second promotion histogram used when mapping data portions to the second of FC storage tier, processing traverses the second promotion histogram, from highest score to lowest score, mapping data portions to the second storage tier until either the capacity limit for the second storage tier is reached or until the response time performance limit for the second storage tier is reached.
For each storage tier, a performance counter is maintained indicating a modeled current I/O processing rate (e.g., total IOPS) and associated modeled response time based on those data portions currently mapped to the storage tier. As described elsewhere herein, performance curves such as illustrated in
In connection with response time performance modeling for a storage tier, as described elsewhere herein with thin devices, the additional I/Os associated with the data portions being added (via mapping) to a storage pool of a particular storage tier may be modeled as being evenly distributed across drives of the storage pool. In the simplified example described herein with only a single storage pool, the modeled storage pool response time is also the modeled storage tier response time. In the event of multiple storage pools in a single tier where all such pools are used by the SG, an embodiment may choose to evenly distribute the added I/O operations across all drives of the storage pool. As described elsewhere herein, a simplifying assumption is that there are no other consumers of the storage tier capacities than those thin devices under device management using the techniques herein. In the event that there are other types of devices having associated data stored on the storage tiers, the amount of storage consumed and the workload of such device may be considered when determining whether capacity and performance limits have been reached. It should be noted that the even distribution modeling as described above may reflect that which is actually performed by the storage tiers and devices therein being evaluated in connection with thin device storage allocation. If an embodiment allocates thin device storage in a different manner, then such modeling should reflect that which is performed in the embodiment.
In a similar manner, a promotion threshold for the second storage tier is determined by performing processing as described above for the first tier with the difference that the processing is performed for the second storage tier until either the capacity limits or response time performance limit of the first zone are reached for the second storage tier. The foregoing capacity limits and response time performance limits vary with each storage tier. Processing that maps data portions to the second storage tier resumes with the second promotion histogram including new second promotion scores for those unmapped data portions from the previous storage tier processing (e.g., those data portions of the first promotion histogram having first promotion scores below the first storage tier promotion threshold). In this manner, data portions which were not mapped to first tier storage are automatically considered for mapping to storage in the next highest tier. At the end of the second storage tier processing for the current zone, the second storage tier promotion threshold is determined.
Referring to
At step 1052, initialization processing is performed. Step 1052 includes initializing a variable, AMT, that keeps track of the amount of storage portions to zero. Step 1052 also includes initializing an index variable, I, to the maximum score (highest bin). In an embodiment herein, there are five thousand bins, so I would be set to five thousand at the step 1054. Of course, other numbers of bins are also possible. Following step 1052 is step 1054 where AMT is incremented by FREQ[I], the amount of data mapped to bin I. Following the step 1054 is step 1056 where an updated modeled tier RT (response time) is determined. At step 1058, a determination is made as to whether any of the capacity limits and/or response time performance limit for the current tier have been exceeded. Step 1058 may include comparing the updated modeled tier RT to the response time performance limit for the current zone and current target promotion tier. Step 1058 may include comparing the current amount of capacity of the target tier consumed via the modeled mapping represented by AMT to the AP capacity limit. As described elsewhere herein, the total capacity consumed across one or more bins may be determined based on the cumulative frequencies of those bins and the amount of allocated storage of the data portions in the foregoing one or more bins. Step 1058 may include comparing the current amount of capacity of the target tier consumed via the modeled mapping represented by AMT to the SG capacity limit such as may be based on the physical drive capacity limits. If it is determined at the test step 1058 that none of the established limits have been exceeded, then control passes from the test step 1058 to a step 1062 where the index variable, I, is decremented. Following the step 1062, control passes back to the step 1054 for another iteration. If any one or more of the foregoing limits are exceeded, step 1058 evaluates to yes and control proceeds to step 1064 where a score threshold is assigned the value of I. Data portions having a score of I or higher are promoted to the highest level of storage. Following the step 1064, processing is complete.
The methodology for determining score values used to map data portions (indicating promotion candidates) to one or more intermediate storage levels may be similar to that described above in connection with the flow chart 1050. In the case of second and third intermediate storage levels in this current embodiment with 3 storage tiers though, processing may be performed with respect to the second promotion histogram. In an embodiment having more than three storage tiers, new promotion scores and an associated new promotion histogram may be computed for a next lower storage tier as may be needed depending on whether a new promotion score is used.
If a same set of promotion scores is used for determining promotion for two storage tiers, the same promotion histogram may be used. For example, consider a case where there are 4 storage tiers—EFD and three storage tiers of rotating disk drives. A first set of promotion scores and a first promotion histogram may be used as described above to determine which data portions are mapped to the EFD tier. The first histogram may be based on first promotion scores having values calculated with weights P1-P12 selected for the particular EFD tier. Next, a second set of promotion scores may be calculated using a second promotion score different from that used in determining the first histogram. The second histogram may be based on second promotion scores having values calculated with new weights P1-P12 selected whereby P7-P12 may be 1 and P1-P6 may be a suitably selected. With reference to
Once promotion threshold processing has completed for the current zone, demotion threshold processing is performed as will now be described.
Referring to
In one embodiment including three storage tiers—EFD, FC rotating disk drives and SATA disk drives—as described above, a first demotion histogram 1100 may be determined for the EFD storage tier. In a manner similar to that as described above for a first set of promotion scores for the EFD tier, a first set of demotion scores may be determined for the EFD storage tier using first demotion scores having weights P1-P12 selected for the particular EFD storage tier. In the example 1100 of
Subsequently a second demotion histogram may be determined using those data portions which have demotion scores from the first histogram less than S1. In other words, those data portions having demotion scores less than S1 are demoted from the EFD storage tier but now a determination may be made as to which storage tier such demoted data portions are located—the FC or the SATA storage tiers. For those data portions demoted from the EFD storage tier, second demotion scores may be determined for use with a second demotion histogram. The second demotion histogram may be based on second demotion scores having weights P1-P12 selected for the second storage tier of FC rotating disk drives in this example. Thus, the second histogram is similar to the first histogram with reference to
In an embodiment, the demotion threshold for a tier may be determined in any suitable manner. For example, an embodiment may select a demotion threshold with respect to demoting a data portion from a storage tier based on the threshold score determined as the promotion threshold for the storage tier. The demotion threshold may be selected as a score that is the same or lower than the promotion threshold. For example, the demotion threshold may be determined using a constant factor by which the promotion threshold for the same storage tier is multiplied. (e.g. promotion threshold for a tier=1.2*demotion threshold for a storage tier). The foregoing may introduce a stationary zone between the promotion and demotion thresholds for a tier where scores falling this stationary zone are neither promoted or demoted with respect to the storage tier. Introduction of the stationary zone may serve as one mechanism that may be included in an embodiment to limit thrashing with respect to repeatedly promoting and then demoting the same data portions having scores which border the promotion or demotion threshold for a storage tier. The demotion threshold may be selected so that it is always equal to or less than the storage capacity for the SG as may be specified in an associated AP.
In an embodiment herein, the processing performed for demoting data portions (extents) may be similar to processing described in connection with
In some embodiments, when a data or storage portion (e.g., an extent) is selected for promotion, only active subportions (e.g., subextents) are promoted while inactive subportions remain at their current storage level. In an embodiment herein, a subportion is considered active if it has been accessed in the previous 4½ days and is considered inactive otherwise. Of course, other appropriate criteria may be used to deem subportions either active or inactive. In some embodiments, when a data portion (e.g., an extent) is selected for demotion, the entire storage portion may be demoted, irrespective of activity level of subportions. In addition, in some embodiments, appropriate mechanism(s) may be provided to reduce the amount of data that is demoted so that more data is maintained on relative faster physical storage devices. Each extent may be evaluated for promotion first as described above and then for demotion if it has not otherwise qualified for promotion. If an extent does not qualify for promotion or demotion, then no data movement is modeled for the extent and subsequently the extent is also not a candidate for data movement with respect to a set of criteria (e.g., capacity limits and performance zone limits) currently being evaluating through modeling using techniques herein. It should be noted that an extent that qualifies for promotion may not then subsequently be a candidate for demotion. Thus, a candidate that qualifies first for promotion may then be removed as a possible demotion candidate.
After processing is performed for the first and second storage tiers to determine promotion and demotion thresholds using capacity limits and the first zone's performance limits, an overall performance metric for the SG using the physical drives of the storage tiers just processed is determined. In one embodiment, this performance metric may be the modeled average response time (RT) for the SG across all storage tiers just processed and may be represented in EQUATION 6 as:
Average RT=(1/Total I/Os per second)*ΣALL_TIERS(RT of tier*I/O operations per second for the tier)
In EQUATION 6, “Total I/Os per second” is the total number or aggregate of I/Os per second across all physical devices of the SG, “ΣALL_TIERS” is the mathematical summation of the product represented by “(RT of tier*I/O operations per second for the tier)”. It should be noted that the “RT of tier” may represent the average response time of physical devices in a particular tier. Additionally, EQUATION 6 may generally be determined with respect to all SGs and devices thereof currently being evaluated using the techniques herein. The foregoing Average RT may serve as an overall metric regarding performance of the entire SG across all storage tiers considered to determine whether the modeled performance using the response time limits for the first zone is preferable over other response time limits of another zone. The foregoing EQUATION 6 is a weighted average response time calculation that considers the number of I/Os with a given response time. Alternatively, an embodiment may compute an average RT including separate weightings related to technology type. It should be noted in connection with computing the average RT for the SG using EQUATION 6, the RT for each storage tier of the SG is utilized. This RT for each storage tier may be the last modeled RT computed during the histogram traversal as a result of performing promotion and demotion threshold determination and modeling the performance of such proposed data movement candidate data portions. It should be noted that if other criteria, such as utilization, are used in addition to or as an alternative to RT, then an embodiment may compute an overall or average metric across all storage tiers similar to as described above with EQUATION 6. For example, if zones of performance limits are defined for utilization limits for the storage tiers, then a metric for computing average utilization across all storage tiers of devices being evaluated may be used to represent the overall performance criteria used in selecting a set of performance limits in combination with capacity limits, and also the associated promotion/demotion thresholds.
In a similar manner as just described for the first set of performance limits of the first zone, processing is also performed for the next zone 2 (e.g., using the second set of performance limits). Thus, promotion thresholds and an average RT using EQUATION 6 are produced as a result of processing in accordance with capacity limits in combination with performance limits of each zone. After each zone is processed for candidate promotion and demotion thresholds, a determination may be made as to whether to stop further evaluating remaining zones. Such a determination may be made by comparing a first value for the average RT determined using EQUATION 6 for a current zone with second value for the average RT determined using EQUATION 6 for the previously processed zone. For example, after determining promotion and demotion thresholds using zone 1 performance limits in combination with capacity limits (zone 1 scenario) and then zone 2 performance limits in combination with capacity limits (zone 2 scenario), the average RT associated with the zonel scenario may be compared to the average RT associated with the zone 2 scenario. If the average RT for zone 2 scenario does not indicate a sufficient or threshold level of improvement over the average RT for zone 1, then no further zones may be evaluated. An embodiment may define a threshold value that represents the minimum amount of improvement expected in order to continue evaluating further zone scenarios (e.g., determining promotion and demotion thresholds using capacity limits and performance limits for subsequently defined zones). An embodiment may determine a difference in metric values obtained for the average RT for the two zone scenarios to be compared. An improvement between zone scenarios may be determined if there is decrease in the average RT (e.g., lower average RT means better overall performance). This decrease may be larger than the threshold in order for a sufficient level of improvement to be determined. Alternatively, an embodiment may set the threshold value to zero so that any decrease in average RT between scenarios is considered sufficient improvement to proceed with evaluating further zone performance limits in combination with capacity limits.
It should be noted that if one of the capacity limits has been exceeded on a preceding iteration of processing for the prior zone, processing using subsequent zones stops. The processing described herein assumes that the lowest storage tier has sufficient capacity to accommodate storage for any data portions not mapped to the other storage tiers.
Referring to
It should be noted that
Once processing as described in
With reference to
For purposes of simplification, examples above considered a single SG. An embodiment may evaluate multiple SGs in combination if they share physical devices or defined pools of devices so that there is a dependency in that they utilize the same data storage resources. Additionally, there may be other consumers of the physical devices beside those under management of an optimizer or other component using the techniques herein for data movement. For example, there may be devices which not under management of such a component performing data movement using techniques herein for any one or more reasons. When considering the performance limits of storage tiers, an embodiment may determine a performance baseline associated with such devices representing the workload of such devices in the system since such devices may be viewed as having consumed or utilized a portion of the allowable performance limits. The performance baseline may be defined as disk utilization or a response time value that a physical storage device or drive would have if the drive only hosted data storage for devices that are not under management by a component using the techniques herein. In one embodiment this may include those portions of thin devices which may not be moved between physical devices such as between storage tiers. An embodiment may determine the baseline performance in any suitable manner for immovable thin devices. For example, an embodiment may determine the data or thick devices included in a thin device pool servicing the thin device and obtain performance data for each such data device in the thin pool. There is an assumption that the embodiment provides for an distribution of workload within pool data devices. Performance data may be obtained for each moveable thin device using the thin device pool where such performance data indicates the thin device workload as distributed over data devices of the thin pool. For each such data device, the workload associated with immovable thin devices may be determined by subtracting the distributed movable thin device workload associated with the data device from the observed workload of the data device. In other words, for a data device, the workload of the data device attributable to the moveable thin device is subtracted from the total workload of the data device. The result of the foregoing is an estimate of the data device workload attributable to non-moveable thin device portions.
In connection with the defined performance or comfort zones described herein, it should be noted that such zones are determined for the particular resource or service that may be consumed or utilized. In a similar manner, zones may be defined and evaluated in connection with other resources or services which are consumed or utilized in the data storage system. For example, zones and performance modeling variations may be modeled in connection with varying the amount of cache where cache limits may be placed on data cached for particular thick or data devices, thin devices, and other entities which consume cache. As another example, zones of performance limits may be specified for varying performance limits related to one or more DAs that service physical data storage devices. In a similar manner as described herein for storage tiers of physical devices, different performance criteria may be specified in terms of performance zones of limits. For example, with respect to DAs, utilization may be used as a performance metric for which comfort zones are defined.
In connection with avoiding thrashing, described herein are several techniques that may be utilized such as related to using weighting of long term and short term metrics (e.g.,
It should be noted that the criteria which is evaluated using techniques herein may include capacity limits and performance limits. The processing performed herein provides for adaptive tier overloading protection by allowing the system to automatically select from different sets or zones of performance limits as system workload changes. The particular performance limit criteria of response time specified for each tier in each zone is only an example of a performance limit criteria that may be used in an embodiment. For example, performance limit criteria may use one or more other metrics other than response time, such as I/O processing rate (e.g., number of I/Os/second), #reads/second, #writes/second, service time, queue waiting time or wait time, length and/or number of wait queues, and the like. These one or more other metrics may be used alone or in combination with response time limits. Furthermore an embodiment may associate a different weighting factor with each of the different metrics included in performance limits specified for a zone. The weights used for each of the different metric may vary with performance zone. Furthermore, the actual metrics may also vary with performance zone. For example, it may be that for a first zone, a particular response time limit is being evaluated and other performance limit criteria is also included for evaluation. This additional performance limit criteria (e.g., an additional metric) may not considered in evaluation with other response time limits of other zones.
Furthermore, the particular overall metric of average response time used to select between evaluated performance zones may vary in an embodiment from what is described herein. For example, an embodiment may use a different metric other than average response time, or may use the average response time metric, alone or in combination with, other overall performance criteria to evaluate and select between performance zone limits. For example, as described elsewhere herein, an embodiment may also use utilization as the performance metric, alone or in combination with, response time. In such an embodiment, comfort zones of utilization values may be specified and an average utilization may be determined across all storage tiers in a manner similar to calculating and using average response time in EQUATION 6. Utilization may also be modeled in a manner similar to response time as described, for example, in connection with
Described above are techniques where performance limits and capacity limits are included in the criteria used to determine when limits of a storage tier have been reached. The above-mentioned criteria may include performance limits alone, or performance limits in combination with capacity limits. Furthermore, the above-mentioned criteria used in connection with comfort zones may include capacity limits alone without performance limits.
As another simpler variation of the techniques herein with the promotion and demotion scores, an embodiment may rather not evaluate multiple alternatives or comfort zones and may rather consider a single set of limits or criteria including limits that may be based on performance limits as described above (e.g., response time) in combination with capacity limits, performance limits alone, or capacity limits of the different storage tiers alone without performance limits in connection with determining data portions such as extents stored in the various storage tiers. Described below is processing for an embodiment using capacity limits alone. However, other above-noted criteria (e.g., performance limits alone or in combination with capacity limits) may be used in connection with such processing. For example, consider an embodiment including three storage tiers—a first tier of SSD or EFD devices and second and third tiers of rotating disk drives. The second tier may be, for example, FC 15K RPM drives and the third tier may be, for example, SATA drives. The three storage tiers may be ranked from first to third, in terms of performance, from highest to lowest accordingly. When evaluating which data portions may be stored on, or moved to, the first highest performing tier of EFD drives, promotion scores for extents as described above may be determined. Values for P1-P12 may be selected based on the expected storage tier workload characteristics and performance characteristics of the EFD drives as described elsewhere herein. For determining which extents may be mapped to the EFD storage tier, values of P7-P12 may be selected, for example, based on the expected I/O sizes for the particular EFD drives. Values for P1-P6 may also be selected, for example, based on the expected amount of write activity to account for the fact that performance of the EFD devices degrades with larger amounts of write activity. In a similar manner, demotion scores may be determined for the extents as described above. A first promotion histogram may be formed as described above in connection with
Referring to
Once the foregoing of
Based on these second promotion scores, a new second promotion histogram as described in
Once promotion processing has completed, demotion threshold processing may be performed in a manner similar to that as described elsewhere herein with the difference that only capacity limits are utilized where applicable.
In a manner similar to that as described for criteria including only capacity limits to determine a mapping between data portions and physical storage locations on different storage tiers, such criteria may alternatively include, for example, performance limits in combination with capacity limits, or performance limits alone.
It should be noted that although the techniques described herein are used with thin devices providing virtual storage provisioning, the techniques herein may also be used in connection with other types of devices such as those not providing virtual provisioning.
The techniques herein may be performed by executing code which is stored on any one or more different forms of computer-readable media. Computer-readable media may include different forms of volatile (e.g., RAM) and non-volatile (e.g., ROM, flash memory, magnetic or optical disks, or tape) storage which may be removable or non-removable.
While the invention has been disclosed in connection with preferred embodiments shown and described in detail, their modifications and improvements thereon will become readily apparent to those skilled in the art. Accordingly, the spirit and scope of the present invention should be limited only by the following claims.
This application is a continuation of U.S. application Ser. No. 13/331,010 filed on Dec. 20, 2011 (pending), which is incorporated by reference herein.
| Number | Name | Date | Kind |
|---|---|---|---|
| 20050102467 | Lam | May 2005 | A1 |
| 20060224844 | Kano | Oct 2006 | A1 |
| 20080288714 | Salomon | Nov 2008 | A1 |
| 20120047287 | Chiu | Feb 2012 | A1 |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 13331010 | Dec 2011 | US |
| Child | 14698047 | US |
