by Mor Harchol-balter, Sungsim Park Y, Mark E. Crovella Y
http://www.lcs.mit.edu/publications/pubs/pdf/MIT-LCS-TR-767.pdf
Add To MetaCart
Abstract:
The Shortest-Remaining-Processing-Time (SRPT) scheduling policy is known to be the op-timal policy for minimizing mean response time, but it is rarely employed in computing systems foranumber of reasons. These reasons include: lack ofknowledge of task size, fear of starvation of the large tasks, concern over pre-emption overhead, and lack of empirical evidence on the per-formance bene ts of switching to SRPT. In this paper we argue that the special characteristics of Web servers and Web workloads make the usual objections to SRPT less persuasive. We start by arguing that it is possible for Web servers to extract task sizes for a large fraction of tasks. We then compare SRPT to an alternative policy { processor sharing (PS) { which we use as an idealization of typical scheduling policies currently used in Web servers. Our comparisons are made both analytically (assuming Poisson arrivals and an empirically-derived le size distribution) and on trace-driven simulations using logs from operating Web servers. With respect to performance, we show that at high server utilization, the SRPT policy can reduce mean waiting time and mean slowdown over PS by well over an order of magnitude.
Citations
|
557
|
Generating Representative Web Workloads for Network and Server Performance Evaluation
– Barford, Crovella
- 1998
|
|
354
|
Web Server Workload Characterization: The Search for Invariants
– Arlitt, Williamson
- 1996
|
|
217
|
Queueing Systems, volume II. Computer Applications
– Kleinrock
- 1976
|
|
153
|
Changes in Web client access patterns: characteristics and caching implications
– Barford, Bestavros, et al.
- 1999
|
|
145
|
Theory of Scheduling
– Conway, Maxwell, et al.
- 1967
|
|
97
|
Heavy-Tailed Probability Distributions in the World Wide Web”. In A Practical Guide to Heavy Tails
– Crovella, Taqqu, et al.
- 1998
|
|
90
|
Web Facts and Fantasy
– Manley, Seltzer
- 1997
|
|
87
|
Flow and stretch metrics for scheduling continuous job streams
– Bender, Chakrabarti, et al.
- 1998
|
|
45
|
The Queue M/G/1 with the Shortest Remaining Processing Time Discipline
– Schrage, Miller
- 1966
|
|
40
|
Performance evaluation of Web proxy cache replacement policies. Performance Evaluation 2000
– Arlitt, Friedrich, et al.
|
|
38
|
Scheduling algorithms
– Karger, Stein, et al.
- 1997
|
|
12
|
Empirical model of WWW document arivals at access links
– Deng
- 1996
|
|
10
|
Performance Implications of Very Large Service-Time Variances
– Heyman
- 2000
|
|
9
|
Impact of non-poisson arrival sequences for call admision algorithms with and without delay
– Feldmann
- 1996
|
|
9
|
On-line algorithms
– Pinedo
- 1995
|
