G. R. Ganger, B. L. Worthington, R. Y. Hou, and Y. N. Patt. Disk subsystem load balancing: Disk striping vs. conventional data placement. In Proceedings of the International Conference on System Sciences, 1993.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....= 64. 0 1000 2000 3000 4000 5000 Web Server Throughput (reqs sec) F L FOR LDF Basic Segm Figure 8: Web server Throughput as function of multithreading degree. Striping unit = 64 KBytes. small striping units are normally associated with better load balancing among the disks [9]. Thus, our LDF replacement policy seems to be the best alternative, since LDF consistently exhibits the best results for all striping unit sizes. Summary. These results clearly demonstrate that FOR and LDF are superior to comparable techniques for a wide range of parameters. The FOR gains ....

G. R. Ganger, B. L. Worthington, R. Hou, and Y. N. Patt. Disk subsystem load balancing: Disk striping vs. conventional data 10 placement. In Proceedings of the 26th Annual Hawaii International Conference on System Sciences, volume I, pages 40-- 49, Hawaii, January 1993.

....function for a station might be changed to f(j) min(j; 4) for j 0. This new reward function is concave, and balancing the load leads to maximizing the total reward. While the objective of resource allocation depends on the application, often the underlying idea is load balancing ( ChRaN89] [GaWHP93]) There is a rich literature on load balancing, and both static and dynamic versions of the load balancing problem have been studied by numerous authors ( LiuSi88] WilRe93] and references therein) Most of this work focuses on algorithms for load balancing, and studies the performance thru ....

G. R. Ganger, B. L. Worthington, R. Y. Hou, and Y. N. Patt, Disk Subsystem Load Balancing: Disk Striping vs. Conventional Data Placement, Proc. of 26 th Hawaii Int. Conf. on System Sciences, Vol. 1, pp. 40-49, 1993.

.... disks by hashing on the logical address [Kim86, Salem86] The performance impact of disk striping has been studied with both open subsystem models [Kim86, Kim91] and closed subsystem models [Salem86] The load balancing benefits of disk striping have been demonstrated with open subsystem models [Livny87, Ganger93a]. The stripe unit size (i.e. the quantity of data mapped onto one physical disk before switching to the next) is an important design parameter that has also been studied with both open subsystem models [Livny87, Reddy89] and closed subsystem models [Chen90] Storage subsystem models have also ....

....mechanical delays associated with disk drives. In 1 The results are usually compared to those produced by the system level simulator driven by the full system level traces. 2 For this reason, I have used open subsystem simulation driven by traces of observed disk activity in my previous work [Ganger93a, Worthington94]. The results match reality in at least one instance. There is no corresponding trace based workload generator for closed subsystem models, which never match the reality of most workloads. 43 this subsection, I compare two of them. The First Come First Served (FCFS) algorithm services requests ....

[Article contains additional citation context not shown here]

G. Ganger, B. Worthington, R. Hou, Y. Patt, "Disk Subsystem Load Balancing: Disk Striping vs. Conventional Data Placement", Hawaii International Conference on System Sciences, January 1993, pp. 40--49.

....the component modules are highly detailed (e.g. the disk module) and the individual components can be configured and interconnected in a variety of ways. DiskSim has been used in a variety of published studies (and several unpublished studies) to understand modern storage subsystem performance [Ganger93a, Worthington94], to understand how storage performance relates to overall system performance [Ganger93, Ganger95, Ganger95a] and to evaluate new storage subsystem architectures [Worthington95a] DiskSim has been validated both as part of a more comprehensive system level model [Ganger93, Ganger95a] and as a ....

....requests to disks in a round robin fashion. Note that the last two schemes do not model real data layouts. In particular, two requests to the same block will often be sent to different devices. However, these data distribution schemes are useful for investigating various load balancing techniques [Ganger93a]. ffl The third String [ Noredun , Shadowed , Parity disk , or Parity rotated ] specifies the redundancy scheme (which is orthogonal to the data distribution scheme) Noredun indicates that no redundancy is employed. 23 Shadowed indicates that one or more replicas of each data disk are ....

[Article contains additional citation context not shown here]

G. Ganger, B. Worthington, R. Hou, Y. Patt, "Disk Subsystem Load Balancing: Disk Striping vs. Conventional Data Placement", Hawaii International Conference on System Sciences, January 1993, pp. 40--49.

....of the component modules are highly detailed (e.g. the disk module) and the individual components can be con gured and interconnected in a variety of ways. DiskSim has been used in a variety of published studies (and several unpublished studies) to understand modern storage subsystem performance [Ganger93a, Worthington94], to understand how storage performance relates to overall system performance [Ganger93, Ganger95, Ganger95a] and to evaluate new storage subsystem architectures [Worthington95a] DiskSim has been validated both as part of a more comprehensive system level model [Ganger93, Ganger95a] and as a ....

....requests to disks in a round robin fashion. Note that the last two schemes do not model real data layouts. In particular, two requests to the same block will often be sent to di erent devices. However, these data distribution schemes are useful for investigating various load balancing techniques [Ganger93a]. The third String [ Noredun , Shadowed , Parity disk , or Parity rotated ] speci es the redundancy scheme (which is orthogonal to the data distribution scheme) Noredun indicates that no redundancy is employed. Shadowed indicates that one or more replicas of each data disk are ....

[Article contains additional citation context not shown here]

G. Ganger, B. Worthington, R. Hou, Y. Patt, \Disk Subsystem Load Balancing: Disk Striping vs. Conventional Data Placement", Hawaii International Conference on System Sciences, January 1993, pp. 40-49.

No context found.

G. R. Ganger, B. L. Worthington, R. Y. Hou, and Y. N. Patt. Disk subsystem load balancing: Disk striping vs. conventional data placement. In Proceedings of the International Conference on System Sciences, 1993.

No context found.

G. R. Ganger, B. L. Worthington, R. Y. Hou, and Y. N. Patt. Disk subsystem load balancing: Disk striping vs. conventional data placement. In Proceedings of the International Conference on System Sciences, 1993.

No context found.

G. R. Ganger, B. L. Worthington, R. Y. Hou, and Y. N. Patt. Disk Subsystem Load Balancing: Disk Striping vs. Conventional Data Placement. In HICSS '93, 1993.

No context found.

G. R. Ganger, B. L. Worthington, R. Y. Hou, and Y. N. Patt. Disk Subsystem Load Balancing: Disk Striping vs. Conventional Data Placement. In HICSS '93, 1993.

No context found.

Ganger, G. R., Worthington, B. L., Hou, R. Y., and Patt, Y. N. (1993). Disk Subsystem Load Balancing: Disk Striping vs. Conventional Data Placement. Proceedings of 26 th Hawaii International Conference on System Sciences. 1 40-49.

No context found.

Ganger, G. R., Worthington, B. L., Hou, R. Y., and Patt, Y. N. (1993). Disk Subsystem Load Balancing: Disk Striping vs. Conventional Data Placement. Proceedings of 26 th Hawaii International Conference on System Sciences. 1 40-49.

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