K. Amiri, G. Gibson, R. Golding. Highly concurrent shared storage. ICDCS, Apr 2000.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....the functionality of the server (e.g. cache coherence, locating data, and servicing disk requests) among the clients. Concurrency control was identified as one of the critical issues in the network attached storage technology because of inherent lack of a central point of coordination [3]. The concurrency control in the Petal [35] virtual disk storage system and the Frangipani [46] file system is achieved using replicated lock servers which utilize Paxos for consistency. Consequently, Disk Paxos is a natural candidate for enabling lock management in network attached storage ....

K. Amiri, G. A. Gibson, R. Golding. Highly concurrent shared storage. In Proceedings of the International Conference on Distributed Computing Systems (ICDCS2000.

....in some group membership protocols) to outright data corruption (as in a na ve timeout based failure detection scheme) Thus, in practice, these systems must conservatively choose a large fail over period, often longer than 30 seconds, which actually causes the clients to time out. The goal of [1] is to allow clients of a storage area network to directly execute RAID encodings across distributed storage devices. This algorithm relies on the ability of clients to accurately detect the failure of storage devices. Moreover, the algorithm in [1] can result in data loss when certain ....

....causes the clients to time out. The goal of [1] is to allow clients of a storage area network to directly execute RAID encodings across distributed storage devices. This algorithm relies on the ability of clients to accurately detect the failure of storage devices. Moreover, the algorithm in [1] can result in data loss when certain combinations of client and device failures occur. In contrast, our algorithm can tolerate the simultaneous crash of all bricks, and it can make progress whenever a majority recover and are able to communicate. Numerous replication protocols use majority ....

Khalil Amiri, Garth A. Gibson, and Richard Golding. Highly concurrent shared storage. In Proceedings of the International Conference on Distributed Computing Systems (ICDCS 2000.

....require fragments from multiple servers. Moreover, the set of fragments must correspond to the same write operation or else the reconstituted data will be incoherent. Examples of distributed storage systems that use erasure coding include Zebra [21] SwiftRAID [35] Intermemory [10] Cheops [4], Myriad [9] and PASIS [59, 60] A challenge that must be confronted in the design of decentralized storage systems is that of partially completed write operations. Write operations in progress and incomplete write operations by clients that crash are both instances of partially completed write ....

....does not provide strong consistency semantics in the face of data redundancy or concurrent updates. Whereas certain failures may result in unclassifiable writes in our protocol, correct operations may lead to unresolvable situations in Ivy (in this way, Ivy is similar to Coda [25] Amiri et al. [4, 5] use a stripe map to communicate to storage nodes the set of other storage nodes that host related stripe units. In the Palladio project [16] stripe maps enables the masking of partial writes by clients. Since client writes employ two phase commit, storage nodes can detect a failure of the ....

[Article contains additional citation context not shown here]

Khalil Amiri, Garth A. Gibson, and Richard Golding. Highly concurrent shared storage. International Conference on Distributed Computing Systems (Taipei, Taiwan, 10--13 April 2000.

....expressed or implied, of the Air Force Research Laboratory or the U.S. Government. Keywords: Decentralized storage, consistency protocol, versioning servers, distributed file systems 1 Introduction Survivable storage systems (e.g. Petal [19] Myriad [7] SwiftRAID [21] PASIS [35] and Cheops [3]) preserve and provide access to data even when a subset of storage nodes fail. The common architecture for such systems spreads data redundantly (either via replication or erasure coding) across a set of decentralized storage nodes. Even when some have failed, the remaining storage nodes can ....

Khalil Amiri, Garth A. Gibson, and Richard Golding. Highly concurrent shared storage. International Conference on Distributed Computing Systems (Taipei, Taiwan, 10--13 April 2000.

....client distribution agent using maps provided by a third party storage mediator. Another system derived from the Swift architecture is Cheops, a striping file system for CMU NASD storage systems [9, 8] The Swift and Cheops work did not directly address atomicity or recovery issues. Amiri et al. [1] show how to preserve read and write atomicity in a shared storage array using RAID striping with parity. This work focuses primarily on safe concurrent accesses to a fixed space of blocks. It does not address file system consistency in the presence of host failures. A number of scalable file ....

K. A. Amiri, G. A. Gibson, and R. Golding. Highly concurrent shared storage. In Proceedings of 20th International Conference on Distributed Computing Systems (1CDCS'2000.

....new map fragments for extending write operations, and support failure atomic object operations, including mirrored striping, truncate remove, and NFS V3 write commitment (commit) The primary challenge is that there is no single serialization point for writes to the shared array. Amiri et al. [1] have recognized this problem and addressed the atomicity issues; their solution handles storage node failures, but not host failures. The Slice map coordinators use an intention logging protocol in the coordinator to preserve atomicity. The protocol is fully implemented and is the subject of a ....

Khalil Amiri, Garth Gibson, and Richard Golding. Highly concurrent shared storage. In Proceedings of the International Conference on Distributed Compuing Systems, April 2000.

....presence of performance heterogeneity. River s distributed queue (DQ) allows consumers and producers to process data at different rates. The DQ relaxes ordering constraints for better performance, similar to our sets. Graduated declustering uses mirrored data to smooth read performance. Abacus [3 5] is a mobile object system. It dynamically balances load by relocating mobile objects based on blackbox monitoring. Abacus focuses on deciding where to locate objects. Our approach is a hybrid function moving, data driven system. Higher level distributed data structures are parameterized to ....

K. Amiri, G. Gibson, and R. Golding. Highly concurrent shared storage. In 20th International Conference on Distributed Computing Systems (ICDCS '00), pages 298--307. IEEE, Apr. 2000.

....on whether a strongly consistent or a more weakly consistent model is used. In the strongly consistent model, the key cost is the additional latency of a write request that must be committed to several global replicas. In the weakly consistent model, or a model using optimistic concurrency [Amiri00], the key metric is the potential for conflicts given the propagation time of requests to the remote sites. Figure 2 considers both of these factors and shows the number of accesses in our traces that might conflict given the across the world latencies introduced in Table 1. Again, the results are ....

....encompass a wider range of data types and semantics. At the block level of storage access, there may be lessons in work on distributed shared memory systems [Mosberger93, Amza99] as well as in mechanisms developed explicitly for storage systems such as various forms of optimistic concurrency [Adya95, Amiri00] or specialized semantics [Burns00, Yu00] applicable to a subset of workloads. Various forms of concurrency control have been studied for many years in the context of database systems [Gray92] object oriented databases [Butterworth91, Lamb91] and in distributed object systems [Birman93, ....

K. Amiri, G. Gibson and R. Golding. Highly concurrent shared storage. Intl. Conference on Distributed Computing Systems, April 2000.

....Run time Run time . automatically design a system . assign data to devices . plan data migration [Alvarez01, Hall01] characterize app . model devices . predict device perf . express QoS [Borowsky97] execute data migration . QoS enforcement . scalability . file and block access [Golding99, Amiri00] automatically configure devices . what to measure . what it means Figure 1: Storage system lifecycle. The components for automatic design and management of storage systems. 3 multiple places and keep it consistent. The ability to adapt the consistency management within the system to the ....

....a number of global replicas may be acceptable. single writer or partitionable if there is only a single writer for a given piece of data, then consistency can be maintained with token schemes that follow a primary replica or schemes such as publish consistency [Burns00] or optimistic methods [Amiri00, Adya95] that assume interference free updates, but provide schemes for rolling back or re applying changes in the rare case that conflicts do occur. If some amount of inconsistency can be tolerated, than schemes for asynchronous updates and mirroring also apply. multiple writers in applications ....

K. Amiri, G. Gibson and R. Golding. Highly concurrent shared storage. Intl. Conference on Distributed Computing Systems, April 2000.

....failure. Shared network storage arrays present their own atomicity and recovery challenges. In Slice, the block service coordinators preserve atomicity of operations involving multiple storage nodes, including mirrored striping, truncate remove, and NFS V3 write commitment (commit) Amiri et al. [1] addresses atomicity and concurrency control issues for shared storage arrays; the Slice coordinator protocol complements [1] with an intention logging protocol for atomic filesystem operations [2] The basic protocol is as follows. At the start of the operation, the proxy sends to the coordinator ....

.... preserve atomicity of operations involving multiple storage nodes, including mirrored striping, truncate remove, and NFS V3 write commitment (commit) Amiri et al. 1] addresses atomicity and concurrency control issues for shared storage arrays; the Slice coordinator protocol complements [1] with an intention logging protocol for atomic filesystem operations [2] The basic protocol is as follows. At the start of the operation, the proxy sends to the coordinator an intention to perform the operation. The coordinator logs the intention to stable storage. When the operation completes, ....

Khalil Amiri, Garth Gibson, and Richard Golding. Highly concurrent shared storage. In Proceedings of the IEEE International Conference on Distributed Computing Systems (ICDCS), April 2000.

No context found.

Khalil Amiri, Garth A. Gibson, and Richard Golding. Highly concurrent shared storage. International Conference on Distributed Computing Systems (Taipei, Taiwan, 10--13 April 2000.

No context found.

Amiri, K., G.A. Gibson, R. Golding, Highly Concurrent Shared Storage, Int. Conf. On Distributed Computing Systems (ICDCS2000.

No context found.

K. Amiri, G. Gibson, and R. Golding. Highly concurrent shared storage. In International Conference on Distributed Computing Systems, April 2000.

....a RAID isolation atomicity object is anchored to each storage server. This object intercepts all reads and writes to the base storage object and verifies the consistency of updates before committing them. The protocols used by this isolation object are highly scalable and are described elsewhere [2]. The cache object keeps an index of a particular object s blocks in the shared cache kept by the ABACUS filesystem process. The RAID 5 object stripes and maintains parity for individual files across sets of storage servers. The storage objects provide flat storage and can be configured to use ....

....a RAID object that provides storage striping and fault tolerance. The RAID object maintains parity on a per file basis, stripes data across multiple storage servers, and is distributed to allow concurrent accesses to shared stripes by clients by using a timestamp based concurrency control protocol [2]. The RAID object can execute at either the client nodes or the storage servers. The object graph used by files for this experiment is shown in Figure 3. The proper placement of the RAID object largely depends on the performance of the network connecting the client to the storage servers. Recall ....

K. Amiri, G. Gibson, and R. Golding. Highly concurrent shared storage. In Proceedings of the 20th International Conference on Distributed Computing Systems, Taipei, Taiwan, Republic of China, Apr. 2000.

....to a RAID object providing storage striping and fault tolerance. The RAID object maintains parity on a per file basis, stripes data across multiple storage servers, and is distributed to allow concurrent accesses to shared stripes by clients by using a timestamp based concurrency control protocol [AGG00] The RAID object can execute at either the client nodes or the storage servers. The object graph used by files for this experiment is shown 14 74.50 14.23 9.04 28.12 19.18 17.62 Write, LAN Write, SAN Degraded read, SAN 0 10 20 30 40 50 60 70 80 Elapsed time (s) At client ....

K. Amiri, G. Gibson, and R. Golding. Highly concurrent shared storage. In Proceedings of the 20th International Conference on Distributed Computing Systems (to appear), April 2000.

No context found.

K. Amiri, G. Gibson, R. Golding. Highly concurrent shared storage. ICDCS, Apr 2000.

No context found.

K. Amiri et al. Highly Concurrent Shared Storage. In Proceedings of the 20th International Conference on Distributed Computing Systems (ICDCS), April 2000.

No context found.

K. Amiri, G. A. Gibson, and R. Golding. Highly concurrent shared storage. In 20th Int. Conf. on Dist. Comp. Sys. (ICDCS), Taipei, Taiwan, April 2000.

No context found.

K. Amiri, G. A. Gibson, and R. Golding. Highly concurrent shared storage. In International Conference on Distributed Computing Systems (ICDCS), 2000.

No context found.

K. Amiri, G. Gibson, and R. Golding, "Highly Concurrent Shared Storage, " in International Conference on Distributed Computing Systems, 2000, pp. 298--307.

No context found.

K. Amiri, G. Gibson, and R. Golding, "Highly Concurrent Shared Storage, " in International Conference on Distributed Computing Systems, 2000, pp. 298--307.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC