Pei Cao, Swee B. Lim, Shivakumar Venkataraman, and John Wilkes, "The TickerTAIP parallel RAID architecture", Proceedings of the 20th Annual International Symposium of Computer Architecture, May 1993, 52-63.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....Because bus resets are unpredictable, this may result in the destruction of state information necessary to complete an operation. Finally, Cao, Lim, Venkataraman, and Wilkes propose another distributed controller architecture, TickerTAIP, designed specifically for RAID level 5 applications [Cao94] User requests are received by originator nodes, centralized control mechanisms that select the appropriate algorithm to be used and then dispatch disk and parity (XOR) work to worker nodes. Each worker node manages a subset of the disks in the array. Similar to the Seagate model, worker nodes ....

.... isolated these details from users by providing an abstract interface which was closer to that found in today s SCSI drives [ANSI91] Similar methods for abstracting the details of disk array operations were recently proposed in the distributed redundant disk array architecture called TickerTAIP [Cao94] In TickerTAIP, the work required to maintain valid data encodings is performed by workers which are distributed throughout the array. To simplify the management of simultaneous actions occurring across the array, TickerTAIP uses a centralized table in which each entry contains a list of ....

[Article contains additional citation context not shown here]

Cao, P., Lim, S. B., Venkataraman, S., and Wilkes, J. "The TickerTAIP parallel RAID architecture." ACM Transactions on Computer Systems 12(3). (August 1994) 236-269.

....of another Representing RAID Operations as Graphs RAIDframe: A Rapid Prototyping Tool for RAID Systems 47 Version 1. 0 6 24 96 Similar methods for abstracting the details of storage operations were recently proposed in the distributed, redundant disk array architecture called TickerTAIP [Cao94]. In TickerTAIP, the work required to maintain valid data encodings is performed by workers which are distributed throughout the array. To simplify managing simultaneous primitive operations occurring across the array, TickerTAIP uses a centralized table in which each entry contains a list of ....

Cao, P., Lim, S. B., Venkataraman, S., and Wilkes, J. "The TickerTAIP parallel RAID architecture." ACM Transactions on Computer Systems, Vol. 12, No. 3. August 1994, pp. 236-269.

....to one or more host computers. Array controllers and disk buses are often duplicated (indicated by the dotted lines in the figure) so that they do not represent a single point of failure [Katzman77, Menon93] The controller functionality can also be distributed amongst the disks of the array [Cao93]. As disks get smaller [Gibson92] the large cables used by SCSI and other bus interfaces become increasingly unattractive. The system sketched in Figure 3b offers an alternative. It uses high bandwidth, bidirectional serial links for disk interconnection. This architecture scales to large arrays ....

....controller in both of the architectures shown in Figure 2 has two disadvantages: it constitutes either a single point of failure or an expensive system resource that must be duplicated, and its performance and connectivity limit the scalability of the array to larger numbers of disks. Cao et al. [Cao93] described a disk array architecture they call TickerTAIP that distributes the controller functionality amongst several loosely coupled controller nodes. Each node controls a relatively small set of disks (one SCSI string, for example) and communicates with the other nodes via a small, dedicated ....

Cao. P, Lim, S.B., Venkataraman, S., and J. Wilkes, "The TickerTAIP Parallel RAID Architecture," Proceedings of the International Symposium of Computer Architecture (ISCA), 1993, pp. 52-63.

.... mirroring or shadowing) and parity (e.g. RAID 5) Both have been known for many years (e.g. Ouchi78] Storage subsystem models have been used to evaluate design issues for both replication based redundancy (e.g. Bitton88, Bitton89, Copeland89, Hsiao90] and parity based redundancy (e.g. [Muntz90, Lee91, Menon91, Holland92, Menon92, Ng92, Cao93, Hou93, Hou93a, Stodolsky93, Treiber94, Chen95]) Comparisons of replication based and parity based redundancy have also relied largely upon storage subsystem models (e.g. Patterson88, Chen91, Hou93, Hou93b, Mourad93] and measurements of prototypes under similar workloads (e.g. Chen90a, Chervenak91] 19 Dynamic Logical to Physical ....

P. Cao, S. Lim, S. Venkataraman, J. Wilkes, "The TickerTAIP Parallel RAID Architecture", IEEE International Symposium on Computer Architecture, May 1993, pp. 52--63.

....a large configuration is not considered. The DataMesh project [10] that is similar to the experimental system of the DR nets, used Transputers as controllers of a parallel disk system to increase I O performance, but they do not provide high reliability. They then proposed TickerTAIP architecture [11] as an enhancement of DataMesh to adopt the RAID parity technique in tolerating a single fault. It distributes functions for controlling the system like the DR nets, but cannot tolerate multiple faults. For scalability, the DR net removes the controller bottleneck by distributing controlling ....

P. Cao, S. B. Lim, S. Venkataraman, and J. Wilkes. The TickerTAIP parallel RAID architecture. In Proc. of the 20th ISCA, pages 52 -- 63, 1993.

....capacity of tens of megabytes and gigabit communication links in the disk drives. IDISK is designed for decision support systems (DDSs) but a disk array with a number of IDISKs can be made. In such a case, each IDISK will have the functions of a conventional array controller. TickerTAIP DataMesh[2][12] has multiple controllers which are connected to each other by an internal network. In [2] it was shown that the system performance is improved by distribution of the parity calculation. DR net differs from these approaches, though, in that it is an architecture where the system scalability, ....

....for decision support systems (DDSs) but a disk array with a number of IDISKs can be made. In such a case, each IDISK will have the functions of a conventional array controller. TickerTAIP DataMesh[2] 12] has multiple controllers which are connected to each other by an internal network. In [2], it was shown that the system performance is improved by distribution of the parity calculation. DR net differs from these approaches, though, in that it is an architecture where the system scalability, considering not only performance but also reliability, is given a high priority. When the ....

P. Cao, S. B. Lim, S. Venkataraman, and J. Wilkes. The TickerTAIP parallel RAID architecture. In Proc. of the 20th ISCA, pages 52 -- 63, 1993.

....from N sources to M sinks. Jain et al. propose graph coloring as a means for optimizing the use of sources and sinks in concurrent transfers[36] It is also possible (Zebra) to utilize multiple servers for a single client, similar to a parallel file system or distributed RAID array. The TickerTAIP [14] distributed array transfers to and from the network attached disks in parallel. 4 NAP Multimedia Research Several research projects concerning using networkattached peripherals in multimedia workstations are ongoing in various universities. The canonical example of the uses for NAPs in ....

....peripherals. The Swift distributed RAID array[12, 44] was the first project to propose striping of data across multiple network connections as an alternative to striping on local disks. Their approach involves creating transfer plans to support the striping. The TickerTAIP distributed RAID array[14] is composed of network attached disks. It represents important work in calculation and management of distributed parity, especially for small writes. As covered in section 3.2, Cray has implemented a Shared File System for a HiPPI RAID array. They have achieved read rates through the file ....

P. Cao, S. B. Lim, S. Venkataraman, and J. Wilkes. The TickerTAIP parallel RAID architecture. In Proc. 20th Annual International Symposium on Computer Architecture, pages 52--63, May 1993.

....see [3, 13, 28] Due to the immensity of sizes and the data rate requirements of multimedia objects, multimedia servers are founded on disk arrays. Disk arrays connect several disks together, and thereby extend the cost, power, and size advantages of small disks to high capacity configurations [4]. A fundamental tradeoff, though, is that large disk arrays are highly susceptible to disk failures [7] To illustrate, although the mean time to failure for a single disk is 300,000 h, an array of 1000 disks will experience a failure every 12 days. Since the storage and retrieval of multimedia ....

....we present failure recovery techniques tailored for video. Such techniques, however, can easily be extended to other data types such as images and animation sequences. 1. 1 Relation to previous work Recently, several research projects have investigated the design of fault tolerant storage systems [4, 18, 33]. Most of these approaches are based on the redundant array of inde 2 P.J. Shenoy, H.M. Vin: Failure recovery algorithms for multimedia servers Fig. 1a,b. Left symmetric and declustered parity organizations in paritybased arrays. M i.j and P i denote data and parity blocks, respectively, and P i ....

Cao P, Lim SB, Venkatraman S, Wilkes J (1993) The TickerTAIP parallel RAID architecture. In: Proceedings of the International Symposium on Computer Architecture, May 1993, San Diego, CA, ACM Press, pp 52--63

....[7, 8] 2 FAULT TOLERANCE Most image and video servers are based on large disk arrays, and hence the ability to tolerate disk failures is central to the design of such servers. The design of fault tolerant storage systems has been a topic of much research and development over the past decade [3, 22]. In most of these systems, fault tolerance is achieved either by disk mirroring [2] or parity encoding [11, 27] Disk mirroring achieves fault tolerance by duplicating data on separate disks (and thereby incurs 100 storage space overhead) Parity encoding, on the other hand, reduces the overhead ....

P. Cao, S. Lim, S. Venkataraman, and J. Wilkes. The TickerTAIP Parallel RAID Architecture. In Proceedings of the Twentieth International Symposium on Computer Architecture, pages 52--63, May 1993.

....for highbandwidth, distributed I O processing on serverless clusters. In early RAID architecture [5] all disks are densely packaged in the same rack with a centralized control. Centralized RAIDs are often attached to a storage server or as a network attached I O subsystem. The TickerTAIP [4] project offered a parallel RAID architecture for supporting parallel disk I O with multiple controllers. Our paper deals with distributed RAID architectures. Distributed RAID concept was explored by Stonebraker and Schloss [27] Prototyping of distributed RAIDs started with the Petal project [19] ....

....and compatibility with current cluster architectures and applications, and (iii) local and remote disk I O operations performed with comparable latency. These requirements Table 1 Research Projects on Parallel and Distributed RAIDs System Attributes USC RAID x [15] Princeton TickerTAIP [4] Digital Petal [19] 29] Berkeley Tertiary Disk [28] RAID architecture environment Orthogonal striping and mirroring over a Linux cluster RAID 5 with multiple controllers as a centralized subsystem Chained declustering in an Unix cluster A RAID 5 built with a Solaris PC cluster Enabling mechanism ....

P. Cao, S. B. Lim, S. Venkataraman, and J. Wilkes, "The TickerTAIP Parallel RAID Architecture", ACM Trans. on Computer System , Vol.12, No.3, August 1994, pp.236-269.

....employed at Compaq Systems Research Center. In this paper, we consider how to design mutual exclusion algorithms under probabilistic assumptions on the timing of steps, rather than hard timing constraints. One motivation for this direction lies in new storage area network (SAN) architectures [1, 2, 7, 11]. The SAN architecture consists of servers sharing a disk farm over a network. Conceptually, with regard to the design of distributed algorithms, the physical model of this architecture is close to the abstract model of shared memory, and consequently shared memory solutions become practical. ....

P. Cao, S. B. Lim, S. Venkatarman, and J. Wilkes. The TickerTAIP Parallel RAID Architecture. ACM Transactions on Computer Systems, 12(3):236-267, Aug. 1994.

....which is shared by all cluster nodes. Each computer or workstation in the cluster may or may not have a local disk. Even if some nodes have attached with local disks, they are accessed locally. The local disk cane be used to buffer data retrieved from the network attached RAID. The TickerTAIP [7], Network attached Secure Disks [13] 14] active disks [34] and Storage Area Network [39] are good examples of network attached storage system [24] 30] 35] This architecture provides direct data transfer between central storage and clients in a networking environment. This architecture is more ....

P. Cao, S. B. Lim, S. Venkataraman, and J. Wilkes, "The TickerTAIP Parallel RAID Architecture", ACM Trans. on Computer System, Vol.12, No.3, August 1994, pp.236-269.

....multiple underlying storage devices. The balancing reduces the hot spot phenomena caused by the ad hoc balancing of a filesystem s namespace, which is familiar from multiple disk mainframe systems [Kim86] Other research on striping systems has emphasized redundancy at the controller level [Cao93] and the management problems that come from incremental growth in storage systems that span multiple storage servers on the network [Lee96, Thekkath97] Server integrated disks (SID) an approach closely related to SAD, exploits the fact that server machines are frequently dedicated resources ....

Cao, P., et al., The TickerTAIP Parallel RAID Architecture, ACM ISCA, May 1993.

....is provided by the file manager to protect the manager s control over storage access policy. The storage mapping metadata is maintained by the drive, allowing smart drives to better exploit detailed knowledge of their own resources to optimize data layout, read ahead, and cache management [Cao94, Patterson95, Golding95]. This is precisely the type of value add opportunity that nimble storage vendors can exploit for market and customer advantage. With mapping metadata at the drive controlling the layout of files, a NASD drive exports a namespace of filelike objects. Because control of naming is more ....

Cao, P., et al., "The TickerTAIP parallel RAID Architecture," ACM Transactions on Computer Systems 12(3). August 1994, 236-269.

....multimedia servers that address these requirements is the subject matter of this paper. 1. 2 Relation to Previous Work During the past few years, we have seen a dramatic increase in the number of academic and industrial research projects investigating the design of fault tolerant storage systems [3, 6, 21]. In most of these systems, fault tolerance is achieved either by disk mirroring [2, 8] or parity encoding [15, 19, 26] Disk mirroring achieves fault tolerance by maintaining duplicate copies of data on separate disks (and thereby incurs 100 storage space overhead) Parity encoding, on the ....

Pei Cao, Swee Boon Lim, Shivakumar Venkatraman, and John Wilkes. The TickerTAIP parallel RAID architecture. In Proceedings of the 1993 International Symposium on Computer Architecture, pages 52--63, May 1993.

....logging parity updates, we have reduced the disk time consumed by parity update by about a factor of eight. 2 1. Our failure model treats disk and controller failures as independent. If concurrent controller and disk failures must be survived, controller state must be partitioned and replicated [Schulze89, Gibson92, Cao93]. 2. Notice that we make no attempt to reduce the cost of the preread and overwrite of the target data block. Additional savings are possible if data writes and deferred and optimally scheduled [Solworth90, Orji93] D T V D TVD TVD TV 3V D T TVD TV D 10 ( 3V T 2 D 10 ( TVD 4 = TVD ....

Cao, P., Lim, S. B., Venkataraman, S., and Wilkes, J. The TickerTAIP Parallel RAID Architecture. In Proceedings of the 20th Annual International Symposium on Computer Architecture (May 1993). IEEE, San Diego, 52-63.

....up. 2.3.2 Storage Optimizations Processing power inside drives and storage subsystems has already been used to optimize functions behind standardized interfaces such as SCSI. This includes optimizations for storage parallelism, bandwidth and access time, including RAID, TickerTAIP, Iceberg [Patterson88,Drapeau94, Wilkes95, Cao94, StorageTek94] and for distributed file system scalability, including Petal, Derived Virtual Devices, and Network Attached Secure Disks [Lee96, VanMeter96, Gibson97] With Active Disks, excess computation power in storage devices is available directly for application specific function in addition to supporting ....

Cao, P., Lim, S.B., Venkataraman, S. and Wilkes, J. "The TickerTAIP Parallel RAID Architecture " ACM Transactions on Computer Systems 12 (3), August 1994.

....of RAID, the marketplace and literature has grown richly with various enhancements and alternative methods of developing disk array architectures. Some examples are the parity striping architecture [27] EMC s Integrated Cached Disk Array (ICDA) 54] the TickerTAIP parallel RAID architecture [14], Network Appliance s NFS File Server Appliance [29] and Iceberg from Storage Technology. In this chapter, the analysis focuses on a RAID 5 level disk array with a cached controller. We extend the idea of restricted overlap to account for the multiple spindles which comprise a RAID 5 device. ....

Cao, P., S.B. Lim, S. Venkataraman, and J. Wilkes. "The TickerTAIP Parallel RAID Architecture". Hewlett-Packard Laboratory Technical Report, (HPL-OSR-93-25), April 1993.

....network ) A related issue here is the fact that in a classic RAID, a single, centralized, controller is used to control the array, and this can be easily become a vulnerable point in the system. We will look in section 5 at a parallel RAID architecture where this problem dissapears (see also [2]) ffl CPU parallelism vs. I O parallelism (data parallelism, SMP, MPP) How can a multiprocessor be connected to a (parallel) RAID What is the best architecture (shared nothing, shared disk) what should be a good cost model that considers the I O cost, the CPU cost and the interconnection ....

....only O(n) operations) The main difficulty lies in performing step 1 in O(log n) time using a linear number of operations. While there is a way to do this, yielding a work optimal list ranking algorithm running in O(log n) see also [6] for details) a simpler, 13 (a) 4 7 3 2 8 9 [3] 1] 1] [2] [1] 0] 4 5 7 3 1 2 6 8 9 [1] 3] 1] 1] 1] 2] 2] 1] 0] x x x [4] 2] 4] b) 6 5 1 Figure 5: Removing nodes of an independent set. a) The initial list. The ranks are given in brackets, and the selected node to be removed are labeled with . b) The list after contraction. ....

[Article contains additional citation context not shown here]

P. Cao, S. B. Lim, S. Venkataraman, and J. Wilkes. 1994. The TickerTAIP Parallel RAID Architecture. ACM Trans. on Computer Systems, Vol. 12, No. 3, Aug. 1994, pp. 236-269.

No context found.

Pei Cao, Swee B. Lim, Shivakumar Venkataraman, and John Wilkes, "The TickerTAIP parallel RAID architecture", Proceedings of the 20th Annual International Symposium of Computer Architecture, May 1993, 52-63.

No context found.

P. Cao, S. B. Lim, S. Venkataraman, and J. Wilkes. The TickerTAIP parallel RAID architecture. In Proc. of the 20th ISCA, pp. 5263, 1993.

No context found.

Pei Cao, Swee Boon Lin, Shivakumar Venkataraman, and John Wilkes. The TickerTAIP parallel RAID architecture. ACM Trans. on Comp. Sys. (TOCS), 12(3):236--269, 1994.

No context found.

Pei Cao, Swee Boo Lim, Shivakumar Venkataraman, and John Wilkes. The TickerTAIP parallel RAID architecture. In Proc. 20th Annual International Symposium on Computer Architecture, pages 52-63, May 1993.

No context found.

P. Cao, S. Lim, S. Venkataraman, and J. Wilkes. The TickerTAIP parallel RAID architecture. ACM Trans. in Computer Systems, 12(3):236--269, Aug. 1994.

No context found.

P. Cao, S. Lim, S. Venkataraman, and J. Wilkes. The TickerTAIP parallel RAID architecture. ACM Transactions on Computer Systems, 12(3):236--269, August 1994.

First 50 documents  Next 50

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