SEARCH
ISTC-CC NEWSLETTER
RESEARCH HIGHLIGHTS
Ling Liu's SC13 paper "Large Graph Processing Without the Overhead" featured by HPCwire.
ISTC-CC provides a listing of useful benchmarks for cloud computing.
Another list highlighting Open Source Software Releases.
Second GraphLab workshop should be even bigger than the first! GraphLab is a new programming framework for graph-style data analytics.
ISTC-CC Abstract
BWS: Balanced Work Stealing for Time-Sharing Multicores
EuroSys'12. Bern, Switzerland, April 10-13, 2012.
Xiaoning Ding*, Kaibo Wang^, Phillip B. Gibbons*, Xiaodong Zhang^
*Intel Labs, Pittsburgh, PA
^Ohio State University
Running multithreaded programs in multicore systems has become a common practice for many application domains. Work stealing is a widely-adopted and effective approach for managing and scheduling the concurrent tasks of such programs. Existing work-stealing schedulers, however, are not effective when multiple applications time-share a single multicore—their management of steal-attempting threads often causes unbalanced system effects that hurt both workload throughput and fairness.
In this paper, we present BWS (BalancedWork Stealing), a work-stealing scheduler for time-sharing multicore systems that leverages new, lightweight operating system support. BWS improves system throughput and fairness via two means. First, it monitors and controls the number of awake, steal-attempting threads for each application, so as to balance the costs (resources consumed in steal attempts) and benefits (available tasks get promptly stolen) of such threads. Second, a steal-attempting thread can yield its core directly to a peer thread with an unfinished task, so as to retain the core for that application and put it to better use. We have implemented a prototype of BWS based on Cilk++, a stateof- the-art work-stealing scheduler. Our performance evaluation with various sets of concurrent applications demonstrates the advantages of BWS over Cilk++, with average system throughput increased by 12.5% and average unfairness decreased from 124% to 20%.
KEYWORDS: work stealing; multicore; time sharing; fairness
FULL PAPER: pdf