M. Harchol-Balter, B. Schroeder, N. Bansal, and M. Agrawal. Implementation of SRPT scheduling in web servers. ACM Transactions on Computer Systems, 21(2):To appear, May 2003.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....be far superior to RR. While most of the literature on scheduling in recent years focuses on fair queueing and round robin (and their weighted variants) there is some work that has a similar philosophy similar to ours: SRPT scheduling is studied, along with its application to web servers, in [2] [3], and the queueing analysis in [1] notes that preemption is required for acceptable performance with long range dependent traffic. The remainder of the paper is organized as follows. Section II introduces our QoS framework and contains some relevant backgrounds. Section III describes the model ....

....at any given time [9] 10] This in turn implies that the average number of transactions, and hence, by Little s Law, the mean response time, is minimized by SRPT. However, SRPT requires knowledge of the transaction length, and may be impractical for many systems (web servers, as considered in [3] are a notable exception) Although our idealized models do assume that all packets for a transaction arrive instantaneously at the bottleneck link when the transaction begins, this is only to avoid modeling factors such as TCP dynamics, and our primary focus is on schedulers that do not require ....

M. Harchol-Balter, N. Bansal, B. Schroeder, M. Agrawal, "Implementation of SRPT Scheduling in Web Servers," Technical Report, CMU-CS-00-170, Carnegie-Mellon University.

.... such as Resource Based Caching consider both the impact of file size and required delivery bandwidth in making cache insertion and replacement decisions [15] Size based scheduling policies at Web servers use the size of the requested file to prioritize connections which request smaller files [3] [8]. Most of these works assume that transmission time primarily depends on the size of the requested file. For instance, 3] 8] use file size as a measure of the size of the connection then use it to optimize their scheduler so as to improve the client experience, or the time taken by the ....

....and replacement decisions [15] Size based scheduling policies at Web servers use the size of the requested file to prioritize connections which request smaller files [3] 8] Most of these works assume that transmission time primarily depends on the size of the requested file. For instance, 3] [8] use file size as a measure of the size of the connection then use it to optimize their scheduler so as to improve the client experience, or the time taken by the client to receive the requested file. However, they do not study the question of whether the size of the requested file is the only ....

M. Harchol-Balter, N. Bansal, and B. Schroeder. Implementation of SRPT Scheduling in Web Servers. Technical Report CMU-CS-00-170, Carnegie Mellon University, 2000.

....of the service obtained thus far. In such situations, SRPT still minimizes the mean response time, but FB need not perform well. Other recent papers that advocate rethinking conventional queueing strategies for heavy tailed traffic include recent work on the use of SRPT in web servers [2] [3] and the theoretical observation in [1] that preemption is re quired for acceptable performance with long range depen dent traffic. The QoS measure we propose in Section III depends on the distribution of the response time, rather than only on its mean. We have not been able to find the optimal ....

M. Harchol-Balter, N. Bansal, B. Schroeder, M. Agrawal, "Im- plementation of SRPT Scheduling in Web Servers," Technical Report, CMU-CS-00-170, Carnegie-Mellon University.

....to be satisfied. In traditional unicast pull dissemination it is well known that the algorithm Shortest Remaining Processing Time (SRPT) optimizes average user perceived latency although current web servers use FIFO out of fear of starving jobs. This may perhaps change as it was recently shown in [25, 6] that all jobs, regardless of length, should prefer SRPT to FIFO if the distribution of lengths is heavy tailed, as is the case when job length is proportional to Web resource size [14] In multicast pull dissemination, it is not too difficult to see that it is impossible for a server to ....

M. Harchol-Balter, N. Bansal, B. Schroeder, and M. Agrawa. Implementation of srpt scheduling in web servers. Technical report, CMU, 2001.

.... response time in such a multi server system, short and long jobs should be assigned to di erent queues [14, 15] Under highly variable workloads, system performance improves considerably if detailed information about the activity of each back end server is available to the front end dispatcher [13]. In general purpose computing systems, this ideal workload partition is hard to achieve because the length of a job is known only a posteriori. In web server systems, however, the size of the le corresponding to a given URL request is a good characterization of the length of the job, so ....

M. Harchol-Balter, N. Bansal, and B. Schroeder. Implementation of SRPT scheduling in web servers. Technical Report CMU-CS-00-170, Carnegie Mellon University, Oct. 2000. 23

....service requirement of the customer being served. The service of the customer that is pre empted is resumed as soon as there are no other customers with a smaller amount of work in the system. Currently, there is renewed interest in the SRPT discipline due to its relevance in Web server modeling [3, 11]. Remark 6.1 Note that if the service requirement distribution has discontinuity points, it may occur (with positive probability) that two customers have the same remaining service requirement, see Schrage and Miller [21] Here we assume this is not the case, thus, B(x) is a continuous function. ....

M. HARCHOL-BALTER, N. BANSAL, B. SCHROEDER. Implementation of SRPT Scheduling in Web Servers. Technical Report CMU-CS-00-170, Carnegie Mellon University (2001).

....methods. In this paper, we are able to obtain a closed form solution using approximations that work well for job sizes possessing the heavy tailed property, which is commonly observed in the Internet measurements. Job size aware scheduling has been implemented in end systems. Recently, in [6], Harchol Balter et al. implement the SRPT scheduling algorithm in a Linux web server. They then argue that since the le distributions in most web servers possess the HT property, such implementation can signi cantly enhance the system performance. However, to observe such enhancement, they have ....

M. Harchol-Balter, N. Bansal, B. Schroeder, and M. Agrawal, \Implementation of SRPT Scheduling in Web Servers," Tech. Rep. CMU-CS-00-170, School of Computer Science, Carnegie Mellon Univ., Pittsburgh, PA, 2001, Submitted for Publication.

....methods. In this paper, we are able to obtain a closed form solution using approximations that work well for job sizes possessing the heavy tailed property, which is commonly observed in the Internet measurements. Job size aware scheduling has been implemented in end systems. Recently, in [6], Harchol Balter ## ### implement the SRPT scheduling algorithm in a Linux web server. They then argue that since the le distributions in most web servers possess the HT property, such implementation can signi cantly enhance the system performance. However, to observe such enhancement, they have ....

M. Harchol-Balter, N. Bansal, B. Schroeder, and M. Agrawal, \Implementation of SRPT Scheduling in Web Servers," Tech. Rep. CMU-CS-00-170, School of Computer Science, Carnegie Mellon Univ., Pittsburgh, PA, 2001, Submitted for Publication.

No context found.

M. Harchol-Balter, B. Schroeder, N. Bansal, and M. Agrawal. Implementation of SRPT scheduling in web servers. ACM Transactions on Computer Systems, 21(2):To appear, May 2003.

No context found.

M. Harchol-Balter, B. Schroeder, N. Bansal, and M. Agrawal. Implementation of SRPT scheduling in web servers. ACM Transactions on Computer Systems, 21(2), May 2003.

No context found.

HarcholBalter, M., Schroeder, B., Bansal, N., and Agrawal, M. 2000. Implementation of SRPT scheduling in web servers. Tech. Rep. CMU-CS-00-170.

No context found.

M. Harchol-Balter, B. Schroeder, N. Bansal, and M. Agrawal. Implementation of SRPT scheduling in web servers. ACM Transactions on Computer Systems, 21(2), May 2003.

No context found.

M. Harchol-Balter, B. Schroeder, N. Bansal, and M. Agrawal. Implementation of SRPT scheduling in web servers. ACM Transactions on Computer Systems, 21(2), May 2003.

No context found.

M. Harchol-Balter, B. Schroeder, N. Bansal, and M. Agrawal. Implementation of SRPT scheduling in web servers. ACM Transactions on Computer Systems, 21(2):To appear, May 2003.

No context found.

M. Harchol-Balter, B. Schroeder, N. Bansal, and M. Agrawal. Implementation of SRPT scheduling in web servers. ACM Transactions on Computer Systems, 21(2):To appear, May 2003.

....FB; LAS; SET; feedback; least attained service; shortest elapsed time; PS; processor sharing; SRPT; shortest remaining processing time; slowdown 1. INTRODUCTION Traditionally the performance of scheduling policies has been measured using mean response time (a.k.a. sojourn time, time in system) [8, 11, 13, 16], and more recently mean slowdown [1, 5, 7] Under these measures, size based policies that give priority to small job sizes (a.k.a. service requirements) at the expense of larger job sizes perform quite well [15] However, these policies tend not to be used in practice due to a fear of ....

....in many applications. For example, in the case of Web servers, it has been shown that by giving priority to requests for small files, a Web server can significantly reduce response times; however it is important that this improvement not come at the cost of unfairness to requests for large files [8]. The same tradeoff applies to other application areas; for example, scheduling in supercomputing centers. Here too it is desirable to get small jobs out quickly, while not penalizing the large jobs, which are typically associated with the important customers. The tradeoff also occurs for age ....

M. Harchol-Balter, B. Schroeder, N. Bansal, and M. Agrawal. Implementation of SRPT scheduling in web servers. ACM Transactions on Computer Systems, 21(2):To appear, May 2003.

No context found.

M. Harchol-Balter et al., "Implementation of SRPT Scheduling in Web Servers", In IPDS 2001.

No context found.

HARCHOL-BALTER, M., BANSAL, N., AND SCHROEDER, B. Implementation of srpt scheduling in web servers. Tech. Rep. CMU-CS-00-170, Carnegie Mellon School of Computer Science, October 2000.

No context found.

M. Harchol-Balter et al., "Implementation of SRPT Scheduling in Web Servers", In IPDS 2001.

No context found.

M. Harchol-Balter, N. Bansal, B. Schroeder, and M. Agrawal, Size-based scheduling to improve web performance, To appear in ACM TOCS. Currently available as CMU Technical Report: CMU-CS-00-170, "Implementation of SRPT Scheduling in Web Servers".

No context found.

M. Harchol-Balter, N. Bansal, B. Schroeder, and M. Agrawal, "Implementation of SRPT scheduling in web servers," Proc. 7th Annual Workshop on Job Scheduling Strategies for Parallel Processing, pp. 11--35, 2001.

No context found.

M. Harchol-Balter et al., "Implementation of SRPT Scheduling in Web Servers", In IPDS 2001.

No context found.

M. Harchol-Balter, N. Bansal, and B. Schroeder. Implementation of SRPT Scheduling in Web Servers. Technical Report, Carnegie Mellon University, CMU-CS-00-170, 2000.

No context found.

M. Harchol-Balter, N. Bansal, B. Schroeder, and M. Agrawal, \Implementation of SRPT Scheduling in Web Servers," Tech. Rep. CMU-CS-00-170, School of Computer Science, Carnegie Mellon Univ., Pittsburgh, PA, 2001. Submitted for Publication.

No context found.

M. Harchol-Balter, N. Bansal, B. Schroeder, and M. Agrawal, \Implementation of SRPT Scheduling in Web Servers," Tech. Rep. CMU-CS-00-170, School of Computer Science, Carnegie Mellon Univ., Pittsburgh, PA, 2001. Submitted for Publication.

First 50 documents

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