Distributed systems, including server farms, web services, and distributed, network-attached storage systems have become increasingly common, providing vast amounts of computational and storage resources. Distributed storage systems use storage space across multiple nodes throughout a potentially wide-area network. Data access routines for accessing data stored on nodes in a distributed storage system must manage multiple client sessions requiring simultaneous access. However, to maintain a consistent view of the data, modification access must be limited to one host at a time. Additionally, a client session performing multiple modifications must maintain data consistency throughout the application of the multiple modifications.
Embodiments of the invention are defined by the claims below, not this summary. A high-level overview of various aspects of the invention are provided here for that reason, to provide an overview of the disclosure, and to introduce a selection of concepts that are further described below in the detailed-description section below. This summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used as an aid in isolation to determine the scope of the claimed subject matter.
Embodiments of the present invention relate to performing multiple modifications to one or more streams as a single atomic unit called an atomic multiple modification. In some embodiments, if one of the multiple modifications is not successful (e.g., if the preconditions are not met), the modifications successfully applied are reversed, returning the streams to their original state before the start of the atomic multiple modification.
Illustrative embodiments of the present invention are described in detail below with reference to the attached drawing figures, wherein:
The subject matter of the present invention is described with specificity herein to meet statutory requirements. However, the description itself is not intended to limit the scope of this patent. Rather, the inventors have contemplated that the claimed subject matter might also be embodied in other ways, to include different steps or combinations of steps similar to the ones described in this document, in conjunction with other present or future technologies. Moreover, although the terms “step” and/or “block” may be used herein to connote different elements of methods employed, the terms should not be interpreted as implying any particular order among or between various steps herein disclosed unless and except when the order of individual steps is explicitly described.
Embodiments of the present invention are directed to making multiple modifications of data that is stored in a distributed manner across a network of storage nodes. In a network composed of multiple nodes distributed throughout a large network, each unit of data can be accessed by a plurality of client sessions, potentially simultaneously. Furthermore, there is consistency to be maintained across different pieces of data across the network. In order for a consistent view of the data to be maintained across all client sessions accessing the data, modification of the data needs to be coordinated. Without coordination over the modification of the data, it is possible for two client sessions to modify two pieces of data in an inconsistent manner, causing unpredictable results. Furthermore, without atomic application of multiple interdependent modifications, it is possible for the failure of a modifying client session to leave data in an inconsistent state, or for a second client session to read data representing an intermediate inconsistent state.
To prevent these types of inconsistencies, primitive operations are required to coordinate the modification of data. In particular, when independent application of individual modifications would create a temporary inconsistent state, it is desirable to have the ability to apply multiple interdependent modifications to one or multiple pieces of data as an atomic transaction, thereby preventing inconsistencies in the view of the data from other client sessions in the distributed network. Furthermore, it is also desirable to be able to cancel an entire set of multiple modifications if one or more of them cannot be successfully applied, thereby preventing inconsistencies in the view of the data resulting from a partially successful transaction. Additionally, it is desirable to be able to specify one or more expected preconditions that must be met before a particular modification can be successfully applied, thereby allowing client sessions to coordinate the application of independent, potentially conflicting transactions in an “optimistic” manner, without necessitating the use of distributed locks.
By way of example, client session X and client session Y are both executing an algorithm that requires them to read the value of data A and compute new values for data A and data B which are based on the previous value of data A and other external data determined by the respective client session (data Cx for client session X and data Cy for client session Y). The new values of A and B can then be written as an atomic operation. However, since no lock is held between reading the value of data A and writing the new values of data A and data B, it is possible that client session X and client session Y both read data A at the same time, then independently compute new values for data A and data B. First client session X and then client session Y will attempt to write their respective new values for data A and data B using an atomic modify transaction, and both will succeed. However, the effect of the modification applied by client session X is effectively undone by the modification applied by client session Y, resulting in potential inconsistency with respect to external data. To prevent this from happening, many traditional network-based file systems would require that client sessions claim a distributed lock across the read of data A through the write of data A and data B, thereby preventing client session Y from reading the value of data A until client session X has completed updating data A and data B. However, use of a distributed lock introduces significant overhead and results in a more fragile distributed system; for example, if client session X fails to release the lock, then client session Y can be prevented from making progress indefinitely. The present invention avoids the need for explicitly distributed locks by allowing client session X and client session Y to specify preconditions that must be met when they attempt to write new values for data A and data B. In this example, client session X and client session Y each require that before data A and data B can be replaced, it must be true that the value of data A has not changed since it was read. Client session Y will succeed in its update, since no modifications occurred between reading data A and writing data A and data B. However, client session X will not succeed in updating any data, since the value of data A changed between reading data A and attempting the update. Client session X is free to abandon the operation, or to restart the entire sequence by reading data A once again. In any case, all data remains consistent both internally and with external state.
The specific data being stored in the network can be in many different forms. Common forms include files and groupings of similar files. According to some embodiments of the present invention, the data is organized into streams, wherein each stream is a data entity composed of miscellaneous stream metadata properties (e.g., a stream name, access control policy, expiration time, etc.), as well as an ordered sequence of references to extents. Each extent is composed of a contiguous ordered sequence of append blocks. Each append block is immutable and is composed of a contiguous ordered sequence of byte values. New extent references may be appended to the end of a stream, but once appended, extent references cannot be deleted from a stream. It is possible for more than one stream to contain references to the same extent, or even for a single stream to contain multiple references to the same extent. New append blocks may be appended to an extent until the extent becomes sealed, after which the extent becomes immutable. Once an extent is sealed, it cannot be unsealed. Append blocks may not be modified or removed from an extent regardless of whether or not the extent is sealed. Therefore, streams and extents only grow in size.
According to one embodiment of the invention, streams are organized in a file system by to storing extent data on a large group of independent, loosely-coupled extent servers, and storing meta-data describing each stream on a centralized server or small, tightly-coupled group of servers. To append data to an extent, the client session interacts with the correct independent extent server; consequently, appending to extents is highly scalable, but does not facilitate coordination of appends to distinct extents. In contrast, the meta-data for all related streams is maintained on a centralized server, so efficient coordination of updates to stream metadata is possible.
By way of example, the meta-data associated with each stream could include information such as the current stream name, a globally unique identifier (GUID), access control policy, expiration time, and an ordered list of references to the extents that make up the stream. One advantage of this organization is that extent references can be copied from one stream to another without the need to move any data in the network. Instead, only reference lists are altered in the meta-data stored on the centralized server. Because the duplication operations only affect stream metadata, they can be efficiently included as part of an atomic multiple modification. In other embodiments of the invention, the meta-data itself could be distributed among multiple nodes in the network. Those skilled in the art will recognize the present invention could apply to data formats other than streams.
According to some embodiments, the invention provides the ability to batch multiple stream precondition constraints and mutation operations into a list of operations and have them execute as one atomic multiple modification. Operations include: stream meta-data precondition constraints, stream creation, stream deletion, stream renaming, duplication of existing extent references from other streams, and setting of miscellaneous stream properties such as expiration time, read-only flag, access control information, and replication policy. Additionally, some embodiments of the invention provide the ability to express cross-stream operations in terms of references to other modification operations within the same multiple modification. For example, within a multiple modification, there may be additional operations applied to streams that are created by the multiple modification. Similarly, within a multiple modification, there may be additional operations applied to streams that are deleted or renamed by the multiple modification operations. By associating each individual operation record in the multiple modification with a target stream identity, and by referring to source streams in a cross-stream operation by their respective record indices in the multiple modification being performed, streams whose identities are created or whose names are changed during the multiple modification can be referenced at any time in the course of the multiple modification. For example if there are 5 records in the list of operations to be performed, then if the 3rd operation is creating a new stream, any of the other operations can refer to that stream as the stream in operation 3.
In accordance with an embodiment, the invention provides the ability to express non-modifying precondition constraints as operations in a multiple modification. These operations by themselves do not result in any change in the state of streams; however, they can cause the entire multiple modification to fail if preconditions are not met. This allows multiple clients to implement coordinated manipulation of stream state without requiring the use of external locks or other synchronization mechanisms. Examples of preconditions that may be specified include: stream existence, stream name, stream identity (a unique ID preserved across rename, but changed across delete/recreate), stream version number, stream expiration time, and extent reference count.
In accordance with a further embodiment, the invention provides the ability to effectively duplicate portions of stream content into new streams as part of an atomic multiple modification. A stream can be modeled as an ordered sequence of references to contiguous data blocks (extents) that are distributed throughout a network of storage nodes. It is possible to copy a stream by simply reading its content and writing a new stream with the same data, but in new extents. However, not only is copying large amounts of data from one extent to another an expensive network/disk-intensive activity that involves multiple network nodes but the same extent may be referenced by more than one stream. It is possible to effectively duplicate very large sections of a stream into a different stream (possibly in a different position) simply by duplicating the extent; this action is lightweight and requires no copying of actual stream data content. Furthermore, all of the state that is manipulated to perform this type of duplication can be maintained by the central meta-data server. A consequence of this model is that duplication of potentially large portions of stream content from one stream into another can be performed as part of an atomic multiple modification, without requiring locks to be held for long periods of time. Clients could use this capability for many purposes, including stream concatenation, garbage collection, log/journal truncation, single-instancing, and aliasing. Additionally, it is possible to batch a large number of unrelated operations together into an atomic multiple modification to reduce the number of transactions required to complete a certain amount of work. Since meta-data modifications are durable transactions, they must be redundantly committed to persistent storage before they can be considered complete. The latency/performance of durable storage (disk) commit operations can be a major limiting factor in the scalability/performance of a transaction system, and the ability to combine many operations into a single commit can be advantageous. Also, since the server can be able to examine all of the operations before applying any changes, optimizations can be implemented that would not be possible otherwise.
In accordance with some embodiments, the invention facilitates the ability to express “all-or-none” semantics and “simultaneous operations.” All-or-none semantics guarantees that if any operation in the multiple modification fails, then the entire multiple modification fails and the state of the streams selected for modification are left in the state prior to the start of the multiple modification. This allows clients to rely on the consistency of the state as well as the atomicity of the multiple modifications. Simultaneous operations in a multiple modification system allow certain sets of modification to occur in a single atomic multiple modification that would be difficult or inefficient if the operations were performed strictly sequentially. By way of example, it is possible in the same multiple modification to rename stream “A” to “B” with one operation, and rename stream “B” to “A” in another operation (i.e., to swap two streams in the namespace with two operations). These two operations would not succeed if performed sequentially, since the first operation would fail due to a namespace collision. Performing the swap sequentially would require three operations and the introduction of a temporary stream name (i.e., rename “A” to “C”, rename “B” to “A”, rename “C” to “B”). It would also complicate the process of “unwinding” a partially completed multiple modification (all-or-none semantics) if one operation fails after others have succeeded. An atomic multiple modification is treated as an unordered set of operations that are all applied simultaneously, eliminating the need for the client to express intermediate consistent states, and allowing the server to optimize implementation of the atomic multiple modification.
According to further embodiments of the invention, the present invention provides the ability to implement simple client-accessible distributed locks and distributed leases using atomic multiple modifications. Stream meta-data includes an expiration time property that can be used to cause a stream to be automatically deleted after a specified interval unless the expiration time is extended. By using the expiration time in conjunction with an atomic multiple modification, a set of clients can compete for a lease or lock by attempting to create a new stream with an agreed-upon name, and can extend a lease by extending the expiration time on the stream that was successfully created. Ownership of a lease/lock may be effectively added as a precondition to any other stream meta-data atomic multiple modification by adding existence of the self-created lease stream and its continued possession of the agreed upon name as additional preconditions to the multiple modification. If a client fails while holding a lease/lock, the lease or lock file can be automatically deleted when its expiration time arrives, allowing another client to claim the lease. Because the lease or lock is represented as an ordinary stream, the tools and mechanisms available for streams (e.g., namespace organization, access control, browsing/viewing, etc.) are intrinsically available for leases and locks as well.
Accordingly, an embodiment of the invention is directed to computer-readable storage media embodying computer-executable instructions for performing a method of performing multiple modifications to one or more streams as a single atomic unit. One or more streams are selected for modification as selected streams. A lock is acquired on the one or more meta-data associated with the selected streams. The names associated with the selected streams are de-coupled from the selected streams. A a plurality of modifications is performed on the selected streams. The names associated with the selected streams are re-coupled to the selected streams. The lock associated with the meta-data associated with the one or more selected streams is released.
According to another embodiment, the invention is directed to computer-readable storage media embodying computer-executable instructions for performing a method of performing multiple modifications to one or more streams as a single atomic unit. One or more streams are selected for modification as selected streams. A lock is acquired on the meta-data associated with the selected streams. The names associated with each of the selected streams are de-coupled from the selected streams. A first modification is performed on the selected streams. It is determined that a second modification would create an inconsistency if performed. The first modification to the selected streams is reversed. The names associated with the selected streams are re-coupled to the selected streams. The lock on the meta-data associated with the selected streams is released.
According to a further embodiment, the invention is directed at computer-readable media storing computer-executable instructions for performing a method of performing multiple modifications to one or more streams as a single atomic unit. One or more append-only streams are selected for modification as selected streams. A lock is acquired on the meta-data associated with the selected streams, where acquiring the lock comprises modifying a stream with a predetermined name to be a lock stream. The name associated with each selected stream is decoupled from its associated selected stream. A first modification is applied to the selected streams. It is determined if a second modification would create an inconsistency if applied. If the modification would create an inconsistency, then the first modification to the selected streams is reversed. If the second modification would not create an inconsistency, then the second modification is applied to the selected streams. The names associated with the selected streams are re-coupled to the selected streams. The lock associated with the meta-data associated with the selected streams is released.
Having described an overview of embodiments of the present invention, an exemplary operating environment in which embodiments of the present invention may be implemented is described below in order to provide a general context for various aspects of the present invention. Referring initially to
By way of example, any of the nodes on the network could act as the location of meta-data describing the streams stored throughout the network, including lists of extent references in each stream, and the set of nodes holding each extent. Client sessions on host 111 and host 106 could simultaneously desire to perform an atomic multiple modification to a stream having extent instances stored on node 112 and node 109. Node 110 could be serving as the meta-data controller. In this case, both hosts 111 and 106 would simultaneously send requests to node 110 to make changes to the desired stream's meta-data and a race would occur. By stipulating appropriate precondition constraints, whichever node originated the first request to be successfully processed by node 110 will be allowed to modify meta-data associated with the stream. The second node's request would fail, requiring the second node to deal with the precondition failure by, for example, abandoning the operation or restarting from a consistent point. This is desired behavior in an optimistic concurrency model. If the first and second nodes need to coordinate more complicated activities that involve multiple distinct transactions, they can establish locks and leases including preconditions on and manipulation of dedicated lock/lease streams in their multiple modification requests, requiring a losing node to wait until the lock is released to perform its modifications.
Referring now to
The invention may be described in the general context of computer code or machine-useable instructions, including computer-executable instructions such as program modules, being executed by a computer or other machine, such as a personal data assistant or other handheld device. Generally, program modules including routines, programs, objects, components, data structures, etc., refer to code that perform particular tasks or implement particular abstract data types. The invention may be practiced in a variety of system configurations, including hand-held devices, consumer electronics, general-purpose computers, more specialty computing devices, etc. The invention may also be practiced in distributed computing environments where tasks are performed by remote-processing devices that are linked through a communications network.
With reference to
Computing device 200 typically includes a variety of computer-readable media. Computer-readable media can be any available media that can be accessed by computing device 200 and includes both volatile and nonvolatile media, removable and non-removable media. By way of example, and not limitation, computer-readable media may comprise computer storage media and communication media. Computer storage media includes both volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer-readable instructions, data structures, program modules or other data. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store the desired information and which can be accessed by computing device 200.
Memory 204 includes computer-storage media in the form of volatile memory. Exemplary hardware devices include solid-state memory, such as RAM. Storage 205 includes computer-storage media in the form of non-volatile memory. The memory may be removable, nonremovable, or a combination thereof. Exemplary hardware devices include solid-state memory, hard drives, optical-disc drives, etc. Computing device 100 includes one or more processors (CPUs) 203 that read data from various entities such as memory 204, storage 205 or I/O devices 206. I/O devices 206 allow computing device 200 to be logically coupled to other devices including input components and output components, some of which may be built in. Illustrative components include a microphone, joystick, game pad, satellite dish, scanner, printer, wireless device, etc. Computing device 200 includes a network device 202, facilitating communication with other network devices. The network device 202 acts as an additional I/O device.
Those skilled in the art will recognize that the data stored in a network of computing devices, called nodes or storage nodes, can be formatted in many ways. By way of example, it is common for independent data files, such as music files, video files, word processing documents, and the like to be stored in a distributed manner. According to an embodiment of the invention, data is modeled as a collection of streams, which may be organized into a conventional hierarchical namespace.
According to some embodiments of the invention, the stream 300 stores references to each extent and/or location or locations of replicas of each extent. For example, a replica of extent 301 is stored on node 304 and instances of extents 302 and 303 are both stored on node 305. It should be noted that extents can be replicated. In those scenarios, the extent data may be accessed at multiple locations, which are the locations of its replicas. The GUID of an extent will not change, but the number and the location of its replicas may change over time. For sake of clarify and simplicity, this example only illustrates the scenario where the extents are not replicated. One advantage of storing references is that it is possible to add an extent reference to a stream without copying or moving any of the data associated with the extent. By way of example, if another copy of extent 301 is to be added to stream 300, the data on node 304 does not need to be altered or copied. Instead, an additional extent reference is added to the list of extent references already in stream 300.
Streams are represented by meta-data, which can itself be distributed throughout the network, or stored on a centralized server, in accordance with an embodiment of the present invention. Referring now to
Atomic multiple modification involves performing multiple modifications to one or more streams. By way of example, modifications include creating a new stream, deleting a stream, renaming a stream, concatenating two or more streams, creating a copy of an existing stream, and copying a subset of the extents of a stream to a second stream. Those skilled in the art will recognize that there are many other possible modifications that could be used in combination to make up an atomic multiple modification. Turning now to
A first modification 509 in the modification set 503 is performed. In this case, stream 501 is to be renamed to “B.” To accomplish this, the name of stream 501 is set to “B”, and stream 501 is reattached to the stream namespace. This is successful without a namespace collision, because the name “B” became available for use when stream 502 was detached from the stream namespace. A second modification 512 from the modification set 503 is also performed. In this case, stream 502 is to be renamed to “A”. To accomplish this, the name of stream 502 is set to “A”, and stream 502 is reattached to the stream namespace. This is successful without a namespace collision, because the name “A” became available for use when stream 501 was detached from the stream namespace.
Referring now to
A lock is acquired on the meta-data associated with each of the one or more streams selected for modification, as well as the stream namespace covered by the multiple modification, as shown at block 602. Once a lock is acquired, the selected streams, which are targeted for deletion or renaming, are de-coupled from their respective names, shown at block 603. According to some embodiments of the invention, this facilitates the application of multiple modifications in a way that the modifications effect the streams as if the modifications occurred simultaneously. By way of example, the decoupling of streams from the associated names is accomplished by creating a temporary mapping storing the original coupling of stream GUIDs to names for each of the selected streams.
Multiple modifications can be made in a way that they occur as an atomic multiple modification. Although two modifications are shown in
Once all of the modifications in the atomic multiple modification are completed, the streams are re-coupled to names, shown at block 606. According to some embodiments of the invention, the stream names are coupled to streams based on changes made to the temporary mapping by each of the modifications applied to the corresponding streams. GUIDs of the streams are used to track these changes and effect the re-coupling. The lock or locks on the meta-data associated with each of the one or more selected streams, as well as the stream namespace covered by the multiple modification is released, as shown at block 607.
Turning now to
Each selected stream is de-coupled from its associated name, as shown at block 703 and a first modification is applied to the one or more selected streams, as shown at block 704, similar to blocks 603 and 604 of
Once it has been determined that a failure will occur in the application of the atomic multiple modification, the modifications that have been applied are reversed, as shown at block 706. After the reversal of any modifications made to the selected streams, streams are re-coupled with their respective original names, and newly created streams are deleted, as shown at block 707. Once the modifications have been reversed, streams have been re-coupled with their original names, and newly created streams have been deleted, the lock is released, as shown at block 708, similar to block 607 in
In accordance with some embodiments of the invention, a method for performing an atomic multiple modification on one or more selected streams is depicted in the flow diagram of
Each stream targeted for deletion or renaming is de-coupled from its associated name, as shown at block 803 and a first modification is performed, as shown at block 804, similar to blocks 603 and 604 of
If it is determined that the second modification can be performed successfully, then the second modification is performed, as shown at block 806, the streams are re-coupled to their final associated names, as shown at block 808, and the lock or locks on the meta-data associated with the streams selected for modification and the stream namespace are released, as shown at block 809. Each of these steps may be performed similarly to steps 605, 606, and 607 of
If, however, it is determined that the second modification would fail, then the first modification is reversed, as shown at block 807, the streams are re-coupled to their original associated names, as shown at block 808, any newly created streams are deleted, and the lock or locks on the meta-data associated with the streams selected for modification and the stream namespace are released, as shown at block 809. Each of these steps may be performed similarly to steps 706, 707, and 708 of
By way of illustration,
For example, consider a system storing a photo album, where the photo album is represented with three streams: ALBUM.CHECKPOINTS 901, ALBUM.DELTA 902, and ALBUM.DATA 903. ALBUM.CHECKPOINTS 901 stores the complete list of all photos, and annotations in the album at a particular point in time. The deletion and addition of photos is stored in ALBUM.DELTA 902. The actual photos are stored in ALBUM.DATA 903. This photo album may be accessed by many thousands of clients concurrently, each adding new photos, editing and replacing photos, deleting photos, renaming photos, annotating photos, etc. Periodically, the ALBUM.DELTA 902 file may become very large and cause clients to experience long load times. Also, extents in ALBUM.DATA 903 may no longer have any references from any recent checkpoint or delta (e.g., the photos in those extents have already been deleted).
Each of the thousands of clients may have each of these streams open. The garbage collection process may eventually decide to remove unneeded items from the streams. For example, assume that ALBUM.CHECKPOINTS 901 consists of two extents, the first with an old checkpoint 904 and the second with the latest checkpoint 905. The old checkpoint could be removed, since the new checkpoint contains a complete system snapshot. ALBUM.DELTA 902 also consists of multiple extents. For this example, assume the relevant deltas from the current checkpoint 905 are all in the last extent 906, of ALBULM.DELTA. ALBUM.DATA 903 contains many extents and scattered among them are extents with no referential streams; in other words, extents that could be removed. For example, only extents with IDs R, S, Y, Z (912, 913, 914, 915) are to be kept in the ALBUM.DATA stream. To clean up the unneeded extents, the garbage collection process could perform the following modifications: remove old extents from ALBUM.CHECKPOINTS 901, remove old extents from ALBUM.DELTA 902, and remove any extents in ALBUM.DATA 903 that are not referenced in the remaining extents in ALBUM.CHECKPOINTS 901 and ALBUM.DELTA 902. The garbage collection process can use an atomic multiple modification to accomplish these modifications.
An example of an atomic multiple modification that the garage collection process could use contains the following modifications. Modification 1: create a new stream named ALBUM.CHECKPOINTS 907, delete stream with STREAM_ID ABCD 901, and append EXTENT_ID B 910 to the new stream, which is assigned ID QRST. Modification 2: create a new stream named ALBUM.DELTA 908, delete stream with STREAM_ID AFDE 902, and append EXTENT_ID L 911 to the new stream, which is assigned STREAM_ID VCDE. Modification 3: create a new stream named ALBUM.DATA 909, delete stream with STREAM_ID DEFF 903, and append EXTENT_ID R 912, EXTENT_ID S 913, EXTENT_ID Y 914, and EXTENT_ID Z 915 to the new stream, which is assigned ID RSTV. Since all modifications are performed atomically, clients never have an inconsistent view of the album. In this particular example, it is also logically correct to modify each of these streams in separate multiple-modification transactions. These streams are modified in a single multiple-modification transaction for efficiency reasons (to reduce the number of transactions). But those skilled in the art will recognize that there are scenarios where multiple streams must be modified in one atomic transaction to ensure consistency among these streams, and multiple-modification support can be used in those scenarios.
Many different arrangements of the various components depicted, as well as components not shown, are possible without departing from the spirit and scope of the present invention. Embodiments of the present invention have been described with the intent to be illustrative rather than restrictive. Alternative embodiments will become apparent to those skilled in the art that do not depart from its scope. A skilled artisan may develop alternative means of implementing the aforementioned improvements without departing from the scope of the present invention.
It will be understood that certain features and subcombinations are of utility and may be employed without reference to other features and subcombinations and are contemplated within the scope of the claims. Not all steps listed in the various figures need be carried out in the specific order described.