Themes COM et NUM | Systemes communicants et Systemes numeriquesEquipe-Projet MoaisRapport de recherche n ° 7077 | version 1 | initial version Octobre 2009 |revised version Novembre 2009 | 21 pagesAbstract:In this paper, we study the problem of dynamic load-balancing on hetero-geneous hierarchical platforms. In particular, we consider here applicationsinvolving heavy communications on a distributed platform. The work-stealingalgorithm introduced by Blumofe and Leiserson is a commonly used techniqueto distribute load in a distributed environment but it su ers from poor per-formances in some cases of communications-intensive applications. We presenthere several variants of this algorithm found in the literature and di erent gridmiddlewares like Satin and Kaapi . In addition, we propose two new variations ofthe work-stealing algorithm : HWS and PWS. These algorithms improve perfor-mances by taking the networking structure into account within the scheduling.We conduct a theoretical analysis of HWS in the case of fork-join task graphsand present experimental results comparing the most relevant algorithms. Ex-periments on Grid'5000 show that HWS and PWS allow us to obtain perfor-mance gains of up to twenty per cent when compared to the standard algorithm.Moreover in some case, the standard algorithm reaches worse performances onthe distributed platform than on a single machine while PWS and HWS achievesome speedup.Key-words: Online scheduling, Work-stealing, Hierarchical platforms
[1]
Henri E. Bal,et al.
Efficient load balancing for wide-area divide-and-conquer applications
,
2001,
PPoPP '01.
[2]
Matteo Frigo,et al.
The implementation of the Cilk-5 multithreaded language
,
1998,
PLDI.
[3]
Thierry Gautier,et al.
KAAPI: A thread scheduling runtime system for data flow computations on cluster of multi-processors
,
2007,
PASCO '07.
[4]
C. Greg Plaxton,et al.
Thread Scheduling for Multiprogrammed Multiprocessors
,
1998,
SPAA.
[5]
Franck Cappello,et al.
Grid'5000: a large scale, reconfigurable, controlable and monitorable Grid platform
,
2005
.
[6]
Jason Maassen,et al.
Adaptive Load Balancing for Divide-and-Conquer Grid Applications
,
2006
.