Spencer Ng. Some design issues of disk arrays. Proceedings of COMPCON Spring '89, pages 137--42. IEEE, 1989. 54  Home/Search   Document Not in Database   Summary   Related Articles   Check

....file implementations File striping is a file implementation that takes advantage of storage device parallelism. These implementations spread, or stripe, the data of a single file across multiple devices (the number of devices is called the stripe width) Files can be striped at the bit level [Tucker88, Ng89], at the byte level [Kim85] or at the block level [Livny87, Patterson88, Henderson89] File striping can potentially multiply throughput by the stripe width [Patterson88] File striping can also decrease queuing time by spreading workload across disks [Livny87] File striping has proven ....

Spencer Ng. Some design issues of disk arrays. Proceedings of COMPCON Spring '89, pages 137--42. IEEE, 1989. 54

....[19] For all of these the maximum data rate is limited by the interconnection medium which is an I O channel. Higher data rates can be achieved by using multiple I O channels. The aggregation of data rates proposed in the Swift architecture generalizes that proposed by the Raid disk array system [3, 8, 20] in its ability to support data rates beyond 13 0 2000000 4000000 6000000 8000000 10000000 12000000 0 5 10 15 20 25 30 35 Number of disks DEC RA82 Fujitsu M2372K Wren V Fujitsu M2351A Fujitsu M2361A IBM 3380K Figure 6: Observed client data rate at maximum sustainable load. ....

S. Ng, "Some design issues of disk arrays," in Proceedings of the IEEE COMPCON Conference, (San Francisco), Feb. 1989.

....distributions at the end of the paper. We call the 70 [ traffic size the peak size. B. Number of Logical Devices (NLD) 15 1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8 File : 8 units long Device 0 Device 1 Device 2 Device 3 Interleaving by 2 units deep R r r r Figure 8: Concept of Depth As reported [14], striping involves the tradeoff of increasing the data transfer rate by a factor of n while reducing the number of logical devices in the storage subsystem by the same factor of n. The NLD or n is calculated by our model using equation (2) Figure 9 shows NLD when peak size is changed from 4[KB] ....

S. Ng, "Some Design Issue of Disk Arrays," Proc. Compcon 89, pp. 137--142, 1989. 46

....consists of multiple disks with data spread over these disks. The ideas of disk interleaving and disk striping were first introduced by Kim [16] and Salem et al. [41] respectively. Since then, a great deal of work has focused on various design issues related to the performance of disk arrays [30, 29, 17, 20, 36, 4, 45] and to their reliability [43, 8, 27] Disk array architectures fall into one of five different classes proposed in [35, 34, 14] referred to as Redundant Arrays of Inexpensive Disks (RAID) Among the five, the two most promising candidates for high performance computing systems appear to be the ....

Ng, S. Some design issues of disk array. In Proc. IEEE Spring ComCon, pages 137--142, 1989.

....form a single logical unit known as a disk array, or RAID (Redundant array of inexpensive disks) 19, 20] In this way, data can be accessed in the array as if it were a single disk volume. There are different RAID levels, each corresponding to how data are laid out and accessed in an application [21, 22, 23]. As a first step performance study on the hierarchical storage system, we characterize the secondary level by its total storage capacity, C 2 , and its total aggregate effective bandwidth, B 2 . The bandwidth B 2 has taken into account the access overheads in the disks (which is related to ....

S. Ng, "Some Design Issues of Disk Arrays," in Digest of Papers: Spring Compcon, pp. 137--142, Feb 27 - Mar 3 1989.

....[23] For all of these the maximum data rate is limited by the interconnection medium which is an I O channel. Higher data rates can be achieved by using multiple I O channels. The aggregation of data rates proposed in the Swift architecture generalizes that proposed by the RAID disk array system [3,9,24] in its ability to support data rates beyond that of the single disk array controller. In fact, Swift can concurrently drive a collection of RAIDs as high speed devices. Due to the distributed nature of Swift, it has the further advantage over RAID of having no single point of failure, such as the ....

S. Ng, "Some design issues of disk arrays," in Proceedings of the IEEE COMPCON Conference, (San Francisco), Feb. 1989.

....reliability by introducing reduncdant data in the system, and discusses the bandwidth issues during R W and rebuild processes. The strengths and weaknesses of different RAID levels in terms of their reliabilities, R W performance, data redundancy, etc. are further examined in [18, 19, 20] Ng in [21] addresses the effect of the stripe size (on system response time and throughput performance) and spindle synchronization (on buffer size) in RAID. Tetzlaff and Flynn review many commercially available video servers or servers under development in terms of their switch architectures (e.g. IBM ....

....a single logical unit known as a disk array, or RAID (Redundant array of inexpensive disks) 16, 18, 37] In this way, data can be accessed in the array as if it were a single disk volume. There are different RAID levels, each corresponding to how data is laid out and accessed in an application [20, 21, 19]. We characterize the secondary level by its total storage capacity, C 2 , and its total aggregate effective bandwidth, B 2 . The bandwidth B 2 has taken into account the access overheads in the disks (which is related to assess latency and block size of the disks) and hence the maximum number ....

S. Ng, "Some design issues of disk arrays," in Digest of Papers: Spring Compcon, pp. 137--142, Feb 27 - Mar 3 1989.

....one disk per request decreases. Except where otherwise noted, all further discussion of disk striping will be in terms of stripe units large enough to ensure that most requests use a single disk. It has also been suggested that striping may be undesirable for transaction processing applications [Ng89] [Gray90] Because these applications are characterized by frequent small disk accesses, it would be undesirable to use more than one disk for any single request. Therefore, one would need to stripe at a very coarse granularity. The arguments made indicate that disk striping is not worth the ....

S. Ng, "Some Design Issues of Disk Arrays", COMPCON, Spring 1989, pp. 137-142.

....= 0; 1 and 3 configurations which we list here. MTTDL0 = 1 m MTTDL1 = 2m 1) 2 m(m 1) 6) MTTDL3 = 6 3 2 (17m 33) 2 (14m 2 47m 33) 2 3 (2m 3 9m 2 11m 3) 4 m(m 1) m 2) m 3) 7) The MTTDL 1 expression has been presented previously by Ng [13]. For reliability modeling such as the above, the ratio = is very large; accordingly we can approximate these mean time to data loss expressions by the following [4] MTTDL c = c m m c c for c = 0; 1; 2; Delta Delta Delta where = The improvement in MTTDL c as a function of c ....

Spencer W. Ng. Some Design Issues of Disk Arrays. In Proceedings of the COMPCOM Conference, pages 137--142, San Francisco, 1989.

No context found.

Spencer Ng, Some Design Issues of Disk Arrays, COMPCON-88, the Thirty-fourth IEEE Computer Society International Conference, March 1989.

No context found.

Spencer Ng, Some Design Issues of Disk Arrays, COMPCON-88, the Thirty-fourth IEEE Computer Society International Conference, March 1989.

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