Drapeau L, Katz R. Striping in Large Tape Libraries. Proceedings of Supercomputing, pp 378387, 1993.  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... I O throughput by reading from multiple tape drives in parallel, it can also increase contention for drives (since each request accesses all drives) Studies have shown that such systems must carefully balance these tradeoffs by choosing an appropriate degree of striping for a given workload [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] ....

A L. Drapeau and R H. Katz. Striping in Large Tape Libraries. In Proceedings of Supercomputing '93, November 1993.

....assumed that requests do not span more than one medium. Though techniques such as striping could be used to split data items across several media, it has been shown that due to the small number of drives in tertiary libraries, striping is not beneficial in the presence of concurrent users [GM95, DK93] If a data item is too large to fit on a single medium, we can split a request for such a large item into separate requests for each contiguous segment. In order to compare the efficiency of potential scheduling policies, it is necessary to define a quantitative measure of performance. This ....

A. L. Drapeau and R. H. Katz. Striping in large tape libraries. In Proc. of Supercomputing, pages 378--387, Portland, Oregon, 1993. ACM. 13

....storage systems. Myllymaki and Livny have investigated the benefits of executing tape and disk I O in parallel [ML96] In [FM96] Ford and Myllymaki have proposed a log structured file system for tertiary media. The benefits of striping in tape based systems has been studied by Drapeau and Katz [DK93] and also by Golubchik and Muntz [GM95] In [GM95] a general open system model for striped libraries is developed. In [Sar95a, Sar95b] modifications to the architecture of database management systems are proposed for efficiently accessing tertiary storage directly. Other studies have looked at ....

A. L. Drapeau and R. H. Katz. Striping in large tape libraries. In Proc. of Supercomputing, pages 378--387, Portland, Oregon, 1993. ACM.

....effective will be the winners in the near line storage market. Today, the chief contenders are high capacity helical scan magnetic tapes and optical disks. This is due mainly to their superior capacity and access time characteristics as compared to conventional tapes, as shown in Table 2 [Ran91] AR93a] Section 6 gives a more complete comparison based on various cost and performance metrics. Table 2: Comparison of Different Storage Technologies. Technology Capacity Data Transfer Average Cost (MB) MBytes sec) Access Time MB Conventional Tape IBM 3480 200 3 15 seconds 0.025 Helical Scan ....

....is that there are few robot arms and a small number of readers (drives) in the systems. Under heavy loads, there is likely to be contention for readers and robot arms when these systems begin to be used for near line purposes. Also, the cost of adding readers to the system may be very high [AR93a] Moderate libraries are less expensive, smaller and often slower, costing around 60,000. Also, in the moderate price range are carousals which hold around 50 cartridges. The SpectraLogix system indicated in Table tapelibraries:tab has the cartridges organized in a carousal. The carousal rotates ....

[Article contains additional citation context not shown here]

Ann L. Drapeau and Randy H. Katz. Striping in Large Tape Libraries. Submitted to Supercomputing Conference., 1993.

.... access through FTP [7] More recently, log structured file systems have been developed that take advantage of the append only and sequential limitations of tapes [6, 3] Drapeau and Katz have studied striping in the context of large tape libraries in the presence of concurrent random I O [2]. They show that in order for striping to be effective in a concurrent environment, it is necessary to have an adequate number of readers. Golubchik and Muntz [4] have studied striping using a more general open system model with multiple sizes of requests within a single run and various stripe ....

A. L. Drapeau and R. H. Katz. Striping in large tape libraries. In Proc. of Supercomputing, pages 378--387, Portland, Oregon, 1993. ACM.

....level. RAID II [CLD 93] views the network as a backplane, with several RAID 5 devices directly connected to a HIPPI network and ethernet. RAID II seeks to deliver high bandwidth for large requests and low latency for small requests. Other projects stripe across networks and tape drives [Sto91, DK93] However, with such low level approaches, there is less user control over buffering and prefetching policies. In addition, it can be argued that software striping can always be used to stripe over a collection of such devices. 5 Future Work and Conclusions We have implemented the Tower of ....

Ann L. Drapeau and Randy H. Katz. Striping in large tape libraries. In Proceedings of Supercomputing '93, pages 378--387, 1993.

....one disk, instead of the whole cache pool like in the RAM. 10 4.6 Jukebox The jukebox is a tape library with 4 tape drives and a robot arm. The tape drive parameters used in this simulation, shown in Table 1, are taken from a paper describing striping in large tape libraries by Drapeau and Katz [1]. The authors did an in depth study of a slow tape library, the Exabyte EXB 120. However, the tape switch time and search time of EXB 120 system is too slow for video on demand use. An average tape switch takes up to 284 seconds. In an environment where tape switches could be frequent, the EXB 120 ....

Ann Drapeau and Randy Katz, "Striping in Large Tape Libraries", Proceedings of Supercomputing `93, pp. 378-387.

....as an archival medium or in a near line system for storing data for which time to first access is not critical. Finally, optical tape has been identified [17] as a promising technology for near line storage systems. Striping techniques for robotic tape libraries are also the focus of much research [9, 8, 14] and offer the potential for improved response time for large data transfers. We note that references [17, 28] contain excellent surveys on storage technologies. In addition, reports on trends in mass storage markets and technologies are available from [22, 21] 2.2.2 High Performance Channels ....

Ann Drapeau and Randy Katz. Striping in Large Tape Libraries. In Proceedings of Supercomputing '93, pages 378--387, Portland, OR, November 1993.

....of striped magnetic tape libraries. They consider a closed simulation model with a workload consisting of fairly small (under 1 GB) reads (75 ) and writes (25 ) of a constant size, within an individual simulation run. The striping configuration is analogous to RAID with block interleaving. In [9], the authors continue their work by considering bit interleaved systems and larger (than 1 GB) requests. Since in this case the striping configuration is analogous to RAID3 [13] the reads are equivalent to K Gamma 1 wide striped requests and the writes are equivalent to K wide striped requests, ....

....4 4 Number of Tapes 116 256 Tape Capacity 5GB 25GB The speeds of the various parts of the storage system, e.g. drives transmission rate, robot loading time, etc. also vary greatly from one system to another. For instance, the following are typical parameters for the Exabyte120, as reported in [9], and the Ampex DST800 [4] Exabyte Ampex Mean Drive Load (sec) 35:4 5 Mean Drive Eject (sec) 16:5 4 Rewind Time ( 1 2 tape) sec) 75 12 13 Search Time ( 1 2 tape) sec) 84 20 30 Transfer Rate (MB sec) 0:47 14:5 Robot Load Unload 22 sec 10 sec The basic operation of servicing an I O request ....

[Article contains additional citation context not shown here]

A. L. Drapeau and R. Katz. Striping in Large Tape Libraries. In Supercomputing '93 Proceedings, pages 378--387, Portland, Oregon, November 1993.

....servicing these requests such that the average waiting time for all requests is minimized. It is assumed that requests do not span more than one medium. Though it is possible that techniques such as striping could be used to split data items across several media, two independent studies [GM95, DK93] have concluded that due to the small number of drives in tertiary libraries, striping is not beneficial in the presence of concurrent users. It is possible that a data item is too large to fit on a single medium, in which case we can split the request into separate requests for each contiguous ....

....[SS93, ML95, Sar95a, Sar95b] Myllymaki and Livny have investigated the benefits of executing tape and disk I O in parallel [ML96] In [FM96] a log structured file system for tertiary media has been proposed. The feasibility of striping in tape based systems has been studied by Drapeau and Katz [DK93] and also by Golubchik and Muntz [GM95] Other studies have looked at the problem of reorganization of data that is stored on tertiary media in order to improve retrieval performance [CDK 95, SS94] 2.3 Problem Specification It has been shown that I O traffic is bursty [RW93] i.e. a large ....

A. L. Drapeau and R. H. Katz. Striping in large tape libraries. In Proc. of Supercomputing, pages 378--387, Portland, Oregon, 1993. ACM.

....[SS93, ML95, Sar95a, Sar95b] Myllymaki and Livny have investigated the benefits of executing tape and disk I O in parallel [ML96] In [FM96] a log structured file system for tertiary media has been proposed. The feasibility of striping in tape based systems has been studied by Drapeau and Katz [DK93] and also by Golubchik and Muntz [GM95] Other studies have looked at the problem of reorganization of data that is stored on tertiary media in order to improve retrieval performance [CDK 95, SS94] In contrast to other studies, we have explored the problem of scheduling requests for more ....

A. L. Drapeau and R. H. Katz. Striping in large tape libraries. In Proc. of Supercomputing, pages 378--387, Portland, Oregon, 1993. ACM.

....transfer of several disks has been extensively studied for RAID [PGK88] architectures. Could a similar technique be employed to improve the low transfer rate of tertiary devices Drapeau and Katz have studied striping in the context of large tape libraries in the presence of concurrent random I O [DK93] They have found that although striping is beneficial for applications whose access patterns consist mainly of large sequential accesses where only one request is handled at a time, it is not necessarily beneficial for concurrent random access patterns. The reason for this poor performance is ....

A. L. Drapeau and R. H. Katz. Striping in large tape libraries. In Proc. of Supercomputing, pages 378--387, Portland, Oregon, 1993. ACM.

....contention for the disks. Instead of generating four independent requests to the disks in the second phase, Separated generates a single coarser request to the fourth disk. This reduces the level of contention and the average queuing delays. Similar effects were also observed by Drapeau and Katz [DK93] in striping experiments with tape libraries. To test this hypothesis, we ran tests where the expansion of the thumbnails was requested one image at a time rather than as a set. We expected that Separated should not perform better than Bundled in this situation. As expected, the crossover point ....

A. L. Drapeau and R. H. Katz. Striping in large tape libraries. In Proc. of Supercomputing, pages 378--387, Portland, Oregon, 1993. ACM.

....and Section 4 presents the corresponding queueing model. In Section 5 we discuss simulation results, and Section 6 describes future work and presents concluding remarks. 2 Related Work Several studies, focusing on extending RAID technology to robotic storage libraries, have been made [10, 11]. In [10] the authors examine the performance of such libraries, where they consider a closed simulation model with a workload consisting of fairly small (under 1 GB) reads (75 ) and writes (25 ) of a constant size, within an individual simulation run. The striping configuration is analogous to ....

....the performance of such libraries, where they consider a closed simulation model with a workload consisting of fairly small (under 1 GB) reads (75 ) and writes (25 ) of a constant size, within an individual simulation run. The striping configuration is analogous to RAID with block interleaving. In [11], the authors continue their work by considering bit interleaved systems and larger (than 1 GB) requests. In this case the striping configuration is analogous to a RAID3 [19] where all files are striped across all the drives in the system. Hence, the reads are equivalent to K Gamma 1 wide ....

[Article contains additional citation context not shown here]

A. L. Drapeau and R. Katz. Striping in Large Tape Libraries. In Supercomputing '93 Proceedings, pages 378--387, Portland, Oregon, November 1993.

....on tape, this distribution is flatter and has less locality than the overall movie request distribution. Because we assume that the distribution is flatter than the overall distribution, we do not simulate replication in the tertiary store. We also do not simulate tape striping; previous work [7] [8] demonstrated that striping is not advantageous for this workload. Tape Libraries in the Hierarchy We evaluate two tape libraries for their performance in a video server storage hierarchy: the relatively low performance, inexpensive Exabyte EXB120 with four tape drives, 116 tapes and a total ....

Ann L. Drapeau and Randy H. Katz. "Striping in large tape libraries." Supercomputing '93, November 1993.

No context found.

Drapeau L, Katz R. Striping in Large Tape Libraries. Proceedings of Supercomputing, pp 378387, 1993.

No context found.

A. Drapeau and R. Katz. "Striping in Large Tape Libraries," Supercomputing `93, November 1993.

No context found.

A. Drapeau and R. Katz. "Striping in Large Tape Libraries," Supercomputing `93, November 1993.

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