USENIX 2002 Annual Conference - Technical Program Abstract
Using Cohort Scheduling to
Enhance Server Performance
James R. Larus and Michael Parkes,
Microsoft Research
Abstract
A server application is commonly organized as a collection of
concurrent threads, each of which executes the code necessary to process a
request. This software architecture, which causes frequent control transfers between
unrelated pieces of code, decreases instruction and data locality, and
consequently reduces the effectiveness of hardware mechanisms such as caches,
TLBs, and branch predictors. Numerous measurements demonstrate this effect in
server applications, which often utilize only a fraction of a modern
processor’s computational throughput.
This paper addresses this problem through cohort scheduling, a new policy that
increases code and data locality by batching the execution of similar
operations arising in different server requests. Effective implementation of
the policy relies on a new programming abstraction, staged computation, which replaces threads. The StagedServer library provides an
efficient implementation of cohort scheduling and staged computation.
Measurements of two server applications written with this library show that
cohort scheduling can improve server throughput by as much as 20%, by reducing the
processor cycles per instruction by 30% and L2 cache misses by 50%.
- View the full text of this paper in
HTML and
PDF.
The Proceedings are published as a collective work, © 2002 by the USENIX Association. All Rights Reserved. Rights
to individual papers remain with the author or the author's employer.
Permission is granted for the noncommercial reproduction of the complete
work for educational or research purposes. USENIX acknowledges all
trademarks within this paper.
- If you need the latest Adobe Acrobat Reader, you can download it from Adobe's site.
- To become a USENIX Member, please see our Membership Information.
|