Optimized disk repository for the storage and retrieval of mostly sequential data

Abstract

A system for the storage and retrieval of data files includes a data disk, a configuration disk and a controller. Each data file includes one or more data blocks, each data block comprising a data section. The data disk stores the data blocks and is divided into a plurality of block groups. The configuration disk stores a copy of at least a portion of the corresponding meta-data. The controller controls both the data disk and the configuration disk.

Description

BACKGROUND

[0001] The present invention is generally related to storage systems. More particularly, the invention is directed to a disk-based storage system which is optimized for storage and retrieval of sequential data.

[0002] Typical disk file systems such as the Microsoft NTFS, Solaris UFS and the Veritas VxFS are designed for the general-purpose storage of data. In particular, disk file systems are architected to handle the simultaneous reading and writing of a large number of both small and large files. To do this efficiently, file systems must implement intricate locking techniques, must be able to cope with random behavior and must be able to perform effectively under a variety of different conditions. This general-purpose design is adequate for some applications but can lead to significant overhead in complexity, performance and capacity.

[0003] In practice, storage intensive applications can be severely limited by general-purpose file systems. This is particularly true for applications that rely on mostly sequential data. Storing and retrieving data in a disk-based backup and data protection environment is an example of a specialized application that suffers from limitations in general-purpose file systems. First, general-purpose file systems typically only scale to less than one terabyte, whereas backup data sets can be significantly larger. Second, general-purpose file systems are fragile, in that they have difficulties in dealing with partial failures. Third, general-purpose file systems handle large streams of data inefficiently. Finally, general-purpose file systems, by their nature, cannot implement special purpose functionality because a feature that may be advantageous for one application may be disastrous for another.

SUMMARY

[0004] The Journaled Object Store (JOS) system of the present invention is a disk repository for the storage of primarily sequential data. The JOS is optimized to handle sequential data, such as is handled by a physical tape drive or a virtual tape drive. The JOS system stores data in two places: 1) a data disk, where all the data together with self-describing meta-data is stored; and 2) a configuration disk, where configuration data and a copy of some of the meta-data are stored. The configuration disk is preferably resident on a non-volatile compact flash disk or other solid-state persistent storage medium. The JOS takes advantage of specialized knowledge of usage patterns to provide levels of reliability, robustness, performance, scalability, availability and data integrity exceeding that of a standard disk file system.

BRIEF DESCRIPTION OF THE DRAWING(S)

[0005] FIG. 1 is Journaled Object Store (JOS) system in accordance with an embodiment of the invention.

[0006] FIG. 2 shows the layout of a data disk in accordance with an embodiment of the invention.

[0007] FIG. 3 shows a data block in accordance with an embodiment of the invention.

[0008] FIG. 4 shows the format of the configuration disk in accordance with an embodiment of the invention.

[0009] FIG. 5 shows the format of a transaction record in accordance with an embodiment of the invention.

[0010] FIG. 6 is a procedure for generating a written transaction record in accordance with an embodiment of the present invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT(S)

[0011] The present invention will be described with reference to the drawing figures where like numerals represent like elements throughout. It should be noted that the terminology “appliance” as used hereinafter refers to any electronic system which implements the present invention, including the controller and the disk. An example of such an appliance is an Intel-based server running virtual tape library software. In this example the JOS is the storage repository for virtual tape and virtual tape library data. In another example, the JOS is the core of a high performance multimedia server for recording and streaming digital media over a network.

[0012] It should also be noted that the term “meta-data” is the data that describes where on a disk, (i.e. on which block regions), a particular data file is stored, along with other aspects of the data file. Meta-data does not include the contents of the data file. Meta-data is essentially data that describes the data file. The terminology “configuration data” is typically associated with system settings and behavior. For example, the number of block groups used by a system could be configuration data, if this number is configurable. The use of the term meta-data herein is intended to encompass both meta-data and configuration data.

[0013] The system of the present invention uses one or more data disks for storing data in the form of data objects. As will be described in detail hereinafter, the layout of the data disk plays an important role in providing reliable and efficient operation of the system.

[0014] Referring to FIG. 1, the JOS system 8 of the present invention comprises a primary unit 10, having a data disk 12 comprising one or more hard disks 12a-12n, a controller 14 and a configuration disk 18. The hard disks 12a-12n could be RAID disk arrays, disk arrays, small computer system interface (SCSI) disks, advanced technology attachment (ATA) disks, Serial ATA, fiber channel (FC) disks, partitions of a single disk and any other present or future developments in storage technology. The configuration disk 18 is preferably a non-volatile flash disk or other non-volatile disk.

[0015] The controller 14 communicates with the data disk 12 and the configuration disk 18 via a communication interface 22 (or bus). As those of skill in the art are aware, there are many types of communication interface technologies that enable data to move between PCs or servers and peripherals, such as hard disk drives. SCSI is a bus architecture which allows multiple devices such as disks, tape drives, libraries, etc. to be daisy-chained. This technology is typically found in high-end storage devices. ATA, which is typically found in PCs, is less costly than SCSI but is also slower for random access and less flexible. Fibre Channel (FC) is a network technology for block-oriented streams, used in storage area networks, which allows for further distances and more complex and flexible topologies.

[0016] The JOS system 8 receives data via an external communication interface 30, (which may be the same or a different type of technology than the communication interface 12), and selectively stores the data and related configuration data into two separate places: 1) the data disk 12; and 2) the configuration disk 18. Both of these storage mediums will be described in further detail hereinafter. It should also be noted that although the data disk 12 and the configuration disk 18 will be described as physically separate components, they may actually be located on the same disk and separated logically via one or more disk partitions.

[0017] Referring to FIG. 2, the layout of the data disk 12 is shown. Each data disk 12 is a logically self-contained storage unit, such as a RAID array, a disk subsystem or an individual disk/partition. The storage capacity is preferably divided into a number of block groups BG1-BGM. A preferred embodiment is shown in FIG. 2 where M block groups are used. As will be explained in detail hereinafter, the use of block groups BG1-BGM helps to provide contiguous allocations of data blocks for storage of data. As such, the number of block groups BG1-BGM can either be fixed, or can be adjusted dynamically depending on external parameters, such as the number of available logical unit number (LUN) groups on the data disk 12 or other practical considerations. A single device can consist of multiple sub-components or units. For example, a disk array can consist of multiple RAID groups. In order to address a specific group on an FC network, an application needs to specify the SCSI ID, (which is the ID of the disk array), along with the LUN that specifies the exact unit within the array. Because each RAID group (LUN) can handle a single disk failure, storing certain information on multiple LUNs greatly improves fault-tolerance. But more importantly, by spreading out disk operations to multiple LUNs, performance is greatly increased. Each block group BG1-BGM includes a disk label block DL1-DLN, a deleted universally unique identifier (Uuid) block DU1-DUN and a plurality of data blocks DB1-DBN.

[0018] The “objects” stored on the data disk 12 comprise data files DF. A data file DF is a logical chain of one or more data blocks DB. Although it is preferable that the chain of data blocks are contiguously located, this is not a requirement. For example, a chain of data blocks DB1-DBn−1 have been graphically highlighted as a data file DF1. A data file DF can span across multiple hard disks 12a-12n, depending on space and bandwidth availability. As will be explained in detail hereinafter, data streams are accepted sequentially and a large area, (i.e., 1 GB or greater), is allocated at a time. Within that area the 1 MB data blocks are generally contiguous. However, since multiple sequential data streams can come in at the same time, it is possible to get slight fragmentation, especially as the data disk 12 gets full or when files are deleted. Accordingly, to further promote locality of access, a plurality of block groups BG1-BGM are used. For example, if four (4) data streams are incoming and four (4) block groups BG1-BG4 are used, each data stream goes to a different block group BG1-BG4.

[0019] A block allocation procedure implemented by the controller 14 allocates different data files DF to different block groups BGs when a request, (i.e., the first “write” operation, or a “create file” operation) is received. This is essentially a request that comes in from an outside entity to place a data file DF onto the data disk 12 (or “object store”). In the case where the system is implemented on a virtual tape library, the request comes in as a write to tape and the system 8 converts it in the appliance to a write to the data disk 12.

[0020] Once the first write occurs, the block allocation procedure prefers to keep additional data blocks DB of a particular data file DF, within the same block group BG to promote locality for efficient access. For example, when a request for storage of the first data block DB1 of the first data file DF1 comes in, it is stored in the first block group BG1. Subsequent data blocks DB2-DBn are forwarded to the first block group BG1 to be sequentially stored with the data file DF1. However, when the first data block DBn+1 of the second data file DF2 comes in, it is preferably stored in the second block group BG2 even though there may be remaining open data blocks in the first block group BG1, (for example, data block n). Subsequent data files DF are sent to different block groups BGs as new data streams are accepted. For example, if five (5) data streams and only four (4) block groups BGs exist, the fifth data stream could be forwarded to the first block group BG1.

[0021] The block allocation procedure applies heuristics when routing data files DF to the data disk 18. First, the block allocation procedure distributes data files DFs equally among block groups BGs, for example in a “round robin” fashion. Alternatively, the block allocation procedure may be based upon the file size and may attempt to use approximately the same amount of space in each block group BG. Second, once the block group BG is determined, the block allocation procedure allocates a large area (such as 1 GB) of memory to the data file DF instead of just a single 1 MB block of memory, and writes the data block at the beginning of this large allocation area. For further data blocks DBs of the same data file DF, the controller 14 writes them contiguously into the same large allocation area. Once the allocation area is full, a new large allocation area of is allocated, ideally contiguously to the first one. The block allocation procedure in accordance with the present invention improves the chances for an efficient layout for specific access pattern; a comparably small number of large sequential reads and writes.

[0022] The disk label block DL1-DLN identifies the disk size and format version information. The disk label DL also contains the appliance identification as well as other parameters that define the disk layout. A data disk 12 with a Securitus I identifier in the disk label DL is henceforth referred to simply as a JOS disk. To assist with data recovery in case of disk label DL damage, the disk label DL1-DLN is preferably repeated at the beginning of each block group BG1-BGM. Possible disk label meta-data and its purpose are shown in Table 1.

[0023] It should be noted that although the meta-data listed in the tables and explained in accordance with the present invention may be utilized by the present invention in a novel manner, the definition of each type of meta-data is well-known in the art.

TABLE 1
Disk Label Meta-Data     Purpose
Device Magic No.         For validation of disk
Version                  Layout version refers to how blocks are
                         laid out on disk. When a disk is imported,
                         it is important to know the version of the
                         layout to be able to find all the data
Device uuid              Uuid for the device
Data Store uuid          For identifying the storage pool, which is a
                         collection of multiple physical disks in one
                         logical pool. Sometimes it is useful to
                         assign specific disks to specific uses or
                         customers to keep the data separate or to
                         route time critical data to a pool of SCSI
                         disks and other data to a pool of ATA
                         disks.
H/W uuid                 For identifying the actual physical
                         hardware
Appliance uuid           For identifying the appliance (i.e., the
                         entire set of hardware and software)
BandWidth                Current bandwidth used
BlockSize                The size of the block
Total blocks on the disk Total number of blocks on the disk
Category                 This flag differentiates between RAID
                         arrays and JBODs, (Just a Bunch Of
                         Disks). In essence, this indicates whether
                         or not the disks are in a fault-tolerant
                         configuration.
Function                 In the case of RAID array, (particularly
                         RAID 3 or 4), it is useful to know which
                         disks are used for data and which disk
                         stores the parity information. In RAID 5,
                         both data as well as parity is stored on all
                         disks.

[0024] In order to make the system more efficient, the disk label block DL1-DLM is written at a known location on each data disk 12. Although it is shown in FIG. 2 as being repeated at the beginning of each block group BG1-BGM, this information may be written to multiple different locations. When a JOS disk gets imported into the system, the first thing the system 8 does is to look for the disk label block DL1-DLM to confirm that it is a JOS disk and to confirm the version of the software used to create the JOS disk. Writing the disk label blocks DL1-DLM to a specific location increases efficiency since the system 8 would otherwise need to scan the entire data disk 12 before finding the necessary disk label blocks DL1-DLM. This location is reserved and does not interfere with any other component of the system 8.

[0025] The deleted Uuid block DU1-DUM records the identity of data files DF that have previously used data blocks DB1-DBN and have later been deleted. For the provision of reliable disaster recovery, the deleted Uuid blocks DU1-DUM are preferably also repeated in each block group BG1-BGM. The deleted Uuid blocks DU1-DUM preferably store the identity of deleted data files DFs that previously used data blocks DBs in that particular block group BG1-BGM. Alternatively, the deleted Uuid blocks DU1-DUM may store the identity of deleted data files DFs (or deleted data blocks DBs) that previously used data blocks DBs in all of the block groups BG1-BGM. In this case, the deleted Uuid blocks DU1-DUM will be identical, and are just repeated in each block group BG1-BGM. Storing the information repeatedly increases fault tolerance at the expense of performance.

[0026] The deleted Uuid blocks DU1-DUM identify blocks which have a consistent header and trailer but belong to data blocks DB1-DBN that have been deleted. During disaster recovery, the information in the header and trailer of data blocks DBs can be used during reconstruction of the configuration of the appliance prior to the last consistent state prior to the crash. The problem is that deleted data blocks DBs on a disk look exactly the same as non-deleted data blocks DBs. Accordingly, during disaster recovery, a data recovery system typically has no way of knowing which data blocks DBs are still active and which data blocks DBs are deleted. However, in accordance with the present invention, the system 8 determines the active data blocks DBs by looking at the deleted Uuid blocks DU1-DUM. The system 8 then knows that only the data blocks DBs that have a header and trailer and are not part of the deleted Uuid blocks DU1-DUM are valid data. Accordingly, the Uuid blocks UB1-UBM are not used to recreate the data, but only to indicate that particular data block DB is no longer important.

[0027] The data blocks DB1-DBN will be explained with reference to FIG. 3. A data block DB1-DBN is the base unit of storage space on the data disk 12. Preferably, each data block DB1-DBN is defined to be 1 MB in size. Such large block sizes are well suited for efficient storage and retrieval of mostly sequential data. However, greater or lesser sizes may be allocated in order to maximize efficiency of the system 8 for a particular application. As shown in FIG. 3, in a preferred embodiment, each data block DB1-DBN includes a header 100 and trailer 120, both of which are embedded within the 1 MB boundary. The header 100 may be extended for variable length data. This provides the system 8 with the ability to handle certain information about files, (the file type, etc.), which could have a variable length. The header 100 and trailer 120 sections verify the integrity of a particular data block DB. The meta-data of the stored data block DB1-DBM is written in the header 100. Preferably, headers 100 and trailers 120 are written in a non-proprietary XML format and contain the device Uuid.

[0028] Different types of meta-data of the data block DB1-DBN are shown in Table 2, along with their primary purpose. Some of the meta-data types (identified as TAPE) are specific to a particular application (i.e., a virtual tape library application), but they can easily be changed to fit other needs by a person skilled in the art. The other meta-data types (identified as JOS) are particular to the system 8 of the present invention. It should be understood that these are examples of types of meta-data that may be utilized by the system 8 of the present invention without departing from the spirit and scope of the present invention.

TABLE 2
Data Block Meta-Data          Purpose
Block Header
JOS Magic No.                 Unique identifier
JOS Data File Uuid            Identifies the file the block is part of
JOS Storage Device Uuid       Identifies the storage device
JOS Volume Uuid               Identifies the volume w/in the storage
                              device
TAPE Appliance Uuid           Identifies the appliance
JOS Volume Pool Uuid          Identifies the volume pool (mult. vols)
TAPE Customer Name            Appliance user information
TAPE Customer Uuid            User identifier
TAPE Library Name             Name of the particular tape library
TAPE Library Uuid             Library identifier
JOS Used                      Size of used blocks
JOS Capacity                  Total available capacity
JOS Bandwidth                 Available bandwidth
JOS Write Protected           Flag for write protection
JOS Label                     Disk Label
TAPE Quick Load               Performance enhancement flag
TAPE Imported                 Indicates if part of appliance
TAPE File Marks               Records tape file marks
TAPE Record Marks             Records tape record marks
JOS Header Length             Indicates the length of the header
JOS File Offset               Indicates the offset w/in the file
JOS DataFile Sequence Change  Ordered sequence number, indicates
Number (SCN)                  where the block belongs in the file
JOS Resource Fork Length      Indicates the length of the header
                              extension
Block Trailer Meta-Data
JOS Data File Uuid            Identifies the file the block is part of
JOS Storage Device Uuid       Identifies the storage device
JOS Header Length             Indicates the length of the header
JOS File Offset               Indicates the offset w/in the file
JOS DataFile Sequence Change  Ordered sequence number, indicates
Number (SCN)                  where the block belongs in the file
JOS Resource Fork Length      Indicates the length of the header
                              extension
JOS Unused bytes in the block Slack space before the trailer

[0029] As explained hereinbefore, the data blocks DB1-DBN are coupled together as a logical chain of blocks to generate a single stream of data associated with a particular data file DF, (i.e., an object). Each single chain of associated data blocks DBs is defined as a data file DF, (for example, DF1), and has a unique identifier (Uuid). The header 100 and trailer 120 of each data block DB1-DBN belonging to a data file DF1 contains the Uuid of the particular data file DF1-DFN. The data file DF1 maintains a monotonically increasing sequence of numbers, called sequence change numbers (SCNs) (or some other ordering scheme), to establish a unique ordering across data blocks DB1-DBN. As each data block DB1-DBn for a data file DF1 is stored, it gets a unique ascending sequence number within that data file DF1. The SCN helps identify the most recent data blocks DB1-DBn even if the SCN within the data file DF1 does not match or for any reason is not consecutive.

[0030] Upon reading a data file DF1, the data blocks DB1-DBn are read back in sequence from the first data block DB1 to the last data block DBn. To the person or application reading the data file DF1 it appears as a single data stream for this single data file DF1. However, in the background, one data block DB1-DBn at a time is read. Of course, “read-ahead” strategies can be employed to make this process even more efficient.

[0031] The secure storage of meta-data is an important aspect of the present invention. In addition to the detailed meta-data stored within the data blocks DB1-DBN, a copy of a subset of the meta-data is also stored centrally for greater reliability and performance. This central repository is the separate configuration disk 18. Preferably, a compact flash disk or flash disk is used as the configuration disk 18. A flask disk is an ideal medium to store all of an appliance's configuration data, as it provides fast and reliable read and write operations. Typically, the reliability of a flash disk is an order of magnitude greater than that of comparable disk media because of the absence of moving parts.

[0032] Preferably, the configuration disk 18 is a self-contained storage unit which provides a circular buffer function; whereby outdated data is eventually overwritten by the latest data. As will be explained in detail hereinafter, data stored on the configuration disk 18 is recorded to the configuration disk 18 in the form of transactional records, (i.e., a self-contained record comprising a header followed by the stored data and a trailer).

[0033] Referring to FIG. 4, a configuration disk (CD) header 300 is written to the configuration disk 18, before any data files DFs are actually written to the data disk 12, and a CD trailer 310 is written to the configuration disk 18 as the last transaction is completed. Transaction records are recorded on the configuration disk 18 as data files DFs are stored on the data disk 12. These transaction records are later used to restore the appliance to its latest consistent state in the event of an unexpected failure or outage. As is well known by those of skill in the art, a consistent state is defined by every transaction being either done or not done, but nothing is half-way done. For example, if a data block DB is being written and the appliance crashes during the write operation, the last known “consistent state” is before that data block DB was written, since it cannot be guaranteed that the last data block DB was correctly written. If a data file DF with many data blocks DBs is being written and, after some of the data blocks DBs are stored the appliance 8 crashes, then the last known consistent state is that the data file DF was not yet written. It should be noted that this is a high level of granularity in that the data file DF is completely written or it is not written at all; it doesn't matter that some of the data blocks DB were successfully written. Of course, this granularity may be changed to the data block DB level if desired.

[0034] In additional to meta-data, the configuration disk 18 may be used to store application-specific data. For example, application-specific data could be the size of a tape library if the system 8 was being used in this context. Alternatively, it could be the number of cars that passed the assembly line if the system 8 was being utilized in a car factory.

[0035] In the embodiment of FIG. 4, the configuration disk 18 uses 512 bytes per data block. The first data block in the configuration disk 18 is the CD header record 300 and the last data block is the CD trailer record 310. In between the CD header and trailer records 300, 310 are data blocks 302-306 that begin with either a TG header record 302 or TG continue record 304 followed by a TG trailer record 306. This facilitates a systematic and efficient method verifying and retrieving meta-data. In this embodiment, the TG continue record 304 is used if the meta-data exceeds 512 bytes (i.e. one data block). The configuration disk 18 is a transaction log implemented as a circular buffer. Each transaction group includes a TG header record 302 and a TG trailer record 306. Once the configuration disk 18 is used up, the very next transaction group is recorded at the beginning of the configuration disk 18. Accordingly, in order for the system 8 to know where to start writing a new transaction group. The CD header record 300 comprises a pointer to the last transaction.

[0036] In operation, the CD header record 300 is read to determine where to start writing in the configuration disk 18. A TG header record 302 is then written. Once the transaction is completed, such as one or more disk writes, a TG trailer record 306 is written and the CD header 300 is updated to the new location, which is now after the latest transaction. The CD trailer record 310 signifies the end of the configuration disk 18. For redundancy, a copy of some or all of the information stored in the CD header record 300 may also be stored there.

[0037] The information stored in the TG header and trailer records 302, 306 is also used during restart. By matching the information in each of the TG header records 302 and the corresponded TG trailer records 306, the system 8 can ensure the data between the TG header and TG trailer records 302, 306 is valid. This is because updates are made sequentially, and the pairing of a TG header record 302 and a TG trailer record 306 therefore ensure standard transactional properties. These standard transactional properties are well known in the art as: atomicity, consistency, isolation and durability, (i.e., ACID properties). During recovery, any TG header record 302 without a corresponding trailer record 306 is assumed to never have happened, (i.e. the transaction is rolled back). The data file DF with the last written TG trailer record 306 is the latest known consistent state.

[0038] Referring to FIG. 5, between the TG header record 302 and the TG trailer record 306 are one or more state stream (SS) records 308. Preferably, these SS records 308 are written in the same data block, unless a TG continue record 304 extends it to the next data block. In the example of FIG. 5 illustrating data on the configuration disk 18, the data block is 512 bytes long. The size of the transaction group is always a multiple of the block size (in the present example N*512 bytes, where N is an integer greater than or equal to one). Each of these SS records 308 contains the updated or new meta-data related to the appliance.

[0039] The essential purpose of the configuration disk 18 is to provide a journaled object store, (hence JOS), or transaction-based, persistent storage repository for configuration and system-wide meta-data (i.e., configuration data that is relevant for the entire JOS, for example for all block groups BGs rather than just one block group BG). Since all write operations are stored as transactions, the configuration disk 18 is extremely robust.

[0040] In the event of an abnormal shutdown, the configuration disk 18 will be accurate up to the point of the last completed transaction. Any transactions that started after that point may not have finished and need to be rolled back. If the TG trailer record 306 has been successfully written to the configuration disk 18, the entire transaction must have been successful. This is because every sub-transaction (i.e., SS record 308) is written sequentially after the TG header record 302. Accordingly, if the header record 302 is present without a matching trailer record 306, it is an indication that some part of the transaction must have gone wrong. When the trailer record 306 is written, it is an acknowledgement, (i.e. a receipt or guarantee), that the meta-data was written to the configuration disk 18 and the actual data to the data disk 12. If this guarantee is not received, the system 8 doesn't know how much of the data file DF was written and, in fact, has to assume that none of the data file DF was written.

[0041] Table 3 sets forth the different types of meta-data identified above along with their associated function.

TABLE 3
Meta-Data   Purpose
Configuration Disk Header and Trailer Records
Magic       Self-identifiable magic number
Uuid        Configuration disk Uuid
Tgid        The expected TransGroup Id to be found
            in the offset in configuration disk
Tid         The expected Transaction Id to be found
            at the offset in configuration disk
Offset      Offset configuration disk location where
            the first and subsequent TransGroup
            records are found
Block Size  Disk block size, which is 512
Size        Size of the configuration disk
Transaction Group Header, Continue and Trailer
Magic       Self-identifiable magic number
TransType   Identifies which of three record types:
            header, continue or trailer
Tgid        Identification for this TransGroup
            record
Uuid        Uuid of the configuration disk
Length      Relative offset to the next TransGroup
            record
Data_length Relative offset to end of the TransGroup
            Trailer record
Transaction Record
Magic       Self-identifiable magic number
Tid         Identification for this Transaction
            record
Length      Relative offset to the next Transaction
            record
State Stream Record
Uuid        Uuid of the configurable object
Status      Active or Deleted
Length      Length of the State Stream record

[0042] The TG header and trailer records 302, 306 comprise two consolidation points which enable the system 8 to restore the appliance to its latest consistent state. During a disaster recovery, the system 8 retrieves all the TG header and trailer records 302, 306 to ensure that they are all paired and to determine which pair is the most current. The CD header 300 contains a pointer to the TG trailer record 306 of the last completed transaction.

[0043] The procedure for generating a written transaction record in accordance with the present invention is shown in FIG. 6. As each data block DB is received by the system 8, a transaction group is stored in the configuration disk 18 as follows: a TG header 302 is written to the configuration disk 18 at the beginning of the next available 512-byte block to initiate the transaction (step 204). Data is written to the data disk 12 one 1-MB block at a time (step 206). For simplicity, this description of the method 200 does not differentiate between data block headers 100, data blocks DB1-DBM and data block trailers 120. For each data block DB, an SS record 308 is written to the configuration disk 18 to indicate the progress or sub-transactions (step 208). If at any time the size of the SS records 308 exceed the size of the 512-byte configuration disk block, the transaction group is extended with a TG continue record 304 (step 210). This could happen multiple times for a large transaction group. Once all SS records 308 are written, (i.e. when the data file DF is completely stored on the data disk 12), a TG trailer record 306 is written to the configuration disk 18 (step 12). This finalizes the transaction and the write is acknowledged.

[0044] As aforementioned, the CD header 300 and trailer 310 are not used to store transactional information. Their function is to identify the currently active region within the configuration disk 18. It should also be noted that the SS records 308 could also identify sub-transactions other than writing a block to the configuration disk 18. The SS records 308 simply denote one entry in the transaction group.

[0045] When the system 8 recovers from a failure, the latest consistent state can be recovered as fine-grained as desired. For example, for an application as a tape library, very coarse grained recovery can be used and the system 8 rolls back to the latest file mark, (i.e., a TG trailer record 306 in the configuration disk 8). To implement a recovery, the system ignores all of the disk transaction records that occurred after the last TG trailer record 306. This is appropriate for backup applications that expect transaction boundaries to be file marks; either the data file DF was backed up or it wasn't. If only parts of the data file DF were backed up it is considered not to be backed up at all.

[0046] The ability of the present invention to handle large sequential data files provides significant advantages over regular file systems.

[0047] With respect to performance, the present invention writes detailed meta-data with each data block DB. Typical prior art data storage systems distinguish between meta-data and regular data, and store meta-data in different data blocks. Since meta-data blocks are kept separate from the actual data blocks, they are typically spread out throughout the disk. This means that every data file write actually involves at least two writes; one that updates the data and at least another one that updates the meta-data. In the system of the present invention, there is only a single write for data and meta-data that is stored together. Since there is no “seeking” of meta-data blocks, a single contiguous write is used.

[0048] Additionally, a reduced set of all of the meta-data within each of the data blocks DBs is also stored together on the configuration disk 18. Writes to the configuration disk 18 are negligible compared to data disk 12 writes because the writes are much smaller and the configuration disk 18 is typically a non-volatile memory that is significantly faster than the data disk 12. The configuration disk 18 keeps a reduced set of the meta-data of each large allocation area, (i.e., the 1 GB blocks referred to hereinbefore), rather than every data block DB. For example, a 40 GB hard drive would require only 40 entries in the configuration disk.

[0049] Another performance enhancement feature of the present invention is related to the large block size. Since the block size is orders of magnitude larger than the block size of ordinary file systems, the advantages of the invention are further magnified. Typical prior art file systems cannot use such large blocks because the block size is the smallest logical unit the file system can write. If many small data files need to be stored, as is typical for an enterprise system, each data file still takes up at least the size of one block. For example, a 1 kb file would take up 1 MB of space, 1023 kb are wasted. Clearly, this is not acceptable for general-purpose systems. In contrast, the system of the present invention is optimized specifically for large sequential data files because backup applications group together these 1 kb files into large, contiguous data files. With these types of access patterns, a large percentage of space will rarely be wasted.

[0050] Prior art disk caches often have to be disabled since it is possible that the disk acknowledges a write before it has propagated from the disk cache to a persistent state. If many small files are written, the file system has to flush the disk cache constantly for consistency before it acknowledges the write to the file. This happens so frequently when many small files are stored that disk caches are simply disabled. In contrast, since the present invention expects a small number of large files, it can make full use of disk caches. Cache flushes become infrequent, (i.e. only at the end of a file to finalize the transaction). This can improve sequential write performance by an order of magnitude.

[0051] Another significant advantage of the present invention is increased reliability in the ability to recover from a corrupted configuration disk 18. Detailed meta-data is written along with each data block, less detailed meta-data into the configuration disk 18. The configuration disk 18 contains a list of the large 1 GB allocation areas that are used to store individual data blocks DBs of a data file DF. One data file DF could cover one or more of these areas. It is not necessary to store detailed meta-data, (for example a list of each data block DB that belongs to a data file DF), into the configuration disk 18 because all of the detailed meta-data is already stored with the data on the data disk 12. In order to access a data file DF, the configuration disk 18 is read to find the first allocation area of the data file DF. Due to this architecture, it is very easy to recover from a corrupt configuration disk 18. The data disk 12 is scanned, and from the detailed meta-data stored in the data blocks DBs, a list of allocation blocks for each data file DF is created again to recover the meta-data of the configuration disk 18. This is not possible in typical prior art file systems because the meta-data is spread throughout the disk. If certain blocks of this meta-data are lost, the disk becomes useless.

[0052] In the same manner, the system of the present invention can easily cope with removable or corrupt disks. If one of the data disks 12 is removed, the data disk 12 contains all of its detailed meta-data. If the disk is imported into another JOS system, it can be scanned and used there. Also, because only meta-data that is specific for data on that particular data disk 12 is removed from the JOS, the system 8 still has all the necessary information to continue running in this degraded state. All data files DFs that are stored on other data disks 12 are guaranteed to still be fully operational. Typical prior art file systems cannot handle such a situation since meta-data is spread over all disks. Accordingly, if one disk is removed, the entire file system is corrupt.

[0053] Finally, with respect to scalability, file systems almost always keep data structures that can address every block on the disk. For example, a 32 bit system with 1 k data blocks can address 232−1*1 Kb (about 2 tera bytes—a small/medium sized tape library). Since the present invention uses 1 MB blocks, it can address several peta bytes, using the same logic, (larger than some of the largest tape libraries). To even further increase this address space, the configuration disk only contains pointers to 1 GB allocation units.

Claims

1. A system for the storage and retrieval of data files, each data file comprising one or more data blocks, each data block comprising a data section and a corresponding meta-data section, the system comprising: a data disk for storing said data blocks; said data disk being divided into a plurality of block groups; a configuration disk for storing a copy of at least a portion of said corresponding meta-data; and a controller for controlling both said data disk and said configuration disk.

2. The system of claim 1 wherein said controller directs for storage of a first data block of a first data file in a first block group of said plurality of block groups and sequentially stores each subsequent data block for the first data file in the first block group.

3. The system of claim 2 wherein said controller directs for storage data blocks of a second data file sequentially in a second block group of said plurality of block groups.

4. The system of claim 2, wherein each data block within said first data file receives a unique sequence number.

5. The system of claim 3, wherein the data blocks within said first data file are each assigned a-sequence number in a continuously ascending sequence.

6. The system of claim 1 wherein said meta-data comprises at least a header and a trailer.

7. The system of claim 1, wherein said meta-data is contained with said data block.

8. A disaster recovery system for the storage and retrieval of data blocks, each data block having a data section and a corresponding meta-data section, the meta-data section comprising a header portion and a trailer portion, the system comprising: a data disk for storing at least one data block, said data disk being divided into a plurality of block groups; a configuration disk, for storing a copy of at least a part of said meta-data in said header portion when storage of a data block in said data disk commences, and storing at least a part of said meta-data in said trailer portion when storage of the data block in said data disk is completed; and a controller, for controlling said data disk and said configuration disk; whereby, upon disaster recovery, said controller matches each said copy of said header portion with said copy of said corresponding trailer portion in said configuration disk to confirm the accurate storage of a data block.

9. A method for the storage and retrieval of data files, each data file comprising one or more data blocks, each data block comprising a data section and a corresponding meta-data section, the system comprising: storing said data blocks on a data disk; said data disk being divided into a plurality of block groups; and storing a copy of at least a portion of said corresponding meta-data on a configuration disk.

10. The method of claim 9 further comprising directing for storage a first data block of a first data file in a first block group of said plurality block groups and sequentially storing each subsequent data block for the first data file in the first block group.

11. The method of claim 10 further comprising directing for storage data blocks of a second data file sequentially in a second block group of said plurality of block groups.

12. The method of claim 10, further comprising assigning each data block within said first data file a unique sequence number.

13. The method of claim 10, further comprising assigning the data blocks within said first data file a sequence number in a continuously ascending sequence.

14. The method of claim 10 wherein said meta-data comprises at least a header and a trailer.

15. The method of claim 1, wherein said meta-data is contained with said data block.

16. A disaster recovery method for the storage and retrieval of data blocks, each data block having a data section and a corresponding meta-data section, the meta-data section comprising a header portion and a trailer portion, the method comprising: storing on a data disk at least one data block, said data disk being divided into a plurality of block groups; storing on a configuration disk a copy of at least a part of said meta-data in said header portion when storage of a data block in said data disk commences, and storing at least a part of said meta-data in said trailer portion when storage of the data block in said data disk is completed; and controlling said data disk and said configuration disk upon disaster recovery to match each said copy of said header portion with said copy of said corresponding trailer portion in said configuration disk to confirm the accurate storage of a data block.

17. A system for the storage and retrieval of data files, each data file comprising one or more data blocks, each data block comprising a data section and a corresponding meta-data section, the system comprising: a plurality of removable data disks for storing said data blocks, each of said data disks being divided into a plurality of block groups; a configuration disk for storing a copy of at least a portion of said corresponding meta-data; and a controller for controlling both of said data disks and said configuration disk; whereby one of said plurality of data disks may be removed without affecting access to the data blocks on other data disks.

18. The system of claim 17 wherein said controller directs for storage of a first data block of a first data file in a first block group of said plurality of block groups and sequentially stores each subsequent data block for the first data file in the first block group.

19. The system of claim 17 further comprising a plurality of configuration disks.

20. The system of claim 19 wherein said plurality of configuration disks may be removed without affecting access to the data blocks on said plurality of data disks.

21. The system of claim 17, wherein said plurality of removable data disks may be dynamically removed and replaced with new removable data disks without affecting operation of the system.