Scheduling Parallel Task Graphs on (Almost) Homogeneous Multicluster Platforms

被引:32
作者
Dutot, Pierre-Francois [1 ]
N'Takpe, Tchimou [2 ]
Suter, Frederic [2 ]
Casanova, Henri [3 ]
机构
[1] Univ Pierre Mendes France, CNRS, INPG,UMR 5217, INRIA,UJF,Grenoble UPMF 1,Grenoble LIG 2, Grenoble 2, France
[2] Nancy Univ, LORIA, CNRS, UMR 7503,INRIA,Nancy UHP 2, Nancy 1, France
[3] Univ Hawaii Manoa, Dept Informat & Comp Sci, Honolulu, HI 96822 USA
基金
美国国家科学基金会;
关键词
Mixed parallelism; parallel task graph scheduling; performance guarantee; multicluster platform; ALGORITHM;
D O I
10.1109/TPDS.2009.11
中图分类号
TP301 [理论、方法];
学科分类号
081202 ;
摘要
Applications structured as parallel task graphs exhibit both data and task parallelism and arise in many domains. Scheduling these applications efficiently on parallel platforms has been a long-standing challenge. In the case of a single homogeneous platform, such as a cluster, results have been obtained both in theory, i.e., guaranteed algorithms, and, in practice, i.e., pragmatic heuristics. Due to task parallelism, these applications are well suited for execution on distributed platforms that span multiple clusters possibly in multiple institutions. However, the only available results in this context are nonguaranteed heuristics. In this paper, we develop a scheduling algorithm, MCGAS, which is applicable to multicluster platforms that are almost homogeneous. Such platforms are often found as large subsets of multicluster platforms. Our novel contribution is that MCGAS computes task allocations so that a (tunable) performance guarantee is provided. Since a performance guarantee does not necessarily imply good average performance in practice, we also compare MCGAS with a recently proposed nonguaranteed algorithm. Using simulation over a wide range of experimental scenarios, we find that MCGAS leads to better average application makespans than its competitor.
引用
收藏
页码:940 / 952
页数:13
相关论文
共 32 条
  • [1] Amdahl G., 1967, Proceedings of AFIPS Spring Join Computer Conference, P483, DOI DOI 10.1145/1465482.1465560
  • [2] [Anonymous], P 15 HET COMP WORKSH
  • [3] [Anonymous], 2009, CPLEX
  • [4] [Anonymous], P 12 INT C PAR DISTR
  • [5] [Anonymous], 1990, Introduction to Algorithms
  • [6] [Anonymous], P 13 ANN ACM S PAR A
  • [7] An improved two-step algorithm for task and data parallel scheduling in distributed memory machines
    Bansal, Savina
    Kumar, Padam
    Singh, Kuldip
    [J]. PARALLEL COMPUTING, 2006, 32 (10) : 759 - 774
  • [8] Grid'5000:: A large scale and highly reconfigurable experimental grid testbed
    Bolze, Raphael
    Cappello, Franck
    Caron, Eddy
    Dayde, Michel
    Desprez, Frederic
    Jeannot, Emmanuel
    Jegou, Yvon
    Lanteri, Stephane
    Leduc, Julien
    Melab, Noredine
    Mornet, Guillaume
    Namyst, Raymond
    Primet, Pascale
    Quetier, Benjamin
    Richard, Olivier
    Talbi, El-Ghazali
    Touche, Irea
    [J]. INTERNATIONAL JOURNAL OF HIGH PERFORMANCE COMPUTING APPLICATIONS, 2006, 20 (04) : 481 - 494
  • [9] An Approximation Algorithm for Scheduling Malleable Tasks under General Precedence Constraints
    Jansen, Klaus
    Zhang, Hu
    [J]. ACM TRANSACTIONS ON ALGORITHMS, 2006, 2 (03) : 416 - 434
  • [10] A NEW POLYNOMIAL-TIME ALGORITHM FOR LINEAR-PROGRAMMING
    KARMARKAR, N
    [J]. COMBINATORICA, 1984, 4 (04) : 373 - 395