网格环境下应用-系统平衡型资源分配优化策略

Application-system balanced optimization for grid resource allocation

针对网格环境中资源分配如何同时优化应用执行性能和提高系统资源效率这个难点问题，提出了基于序贯博弈的执行时间最优目标下的Makespan-效率平衡型优化策略以及费用最优目标下的Cost-效率平衡型扩展策略。优化策略利用平衡的思想将资源分配过程分为活动分发和处理器分配两个阶段，通过有限次博弈后获得优化解。仿真实验以实际网格系统和应用参数为源数据，用典型网格测试床检验了优化策略的可行性和有效性。实验结果表明，相比Min-min算法，优化策略具有更低的时间复杂度、更优的应用执行性能和资源效率。优化策略将活动分发到少数“优质”站点，这种集中性使不准确预测信息带来的Makespan偏差加大。

A novel Application-System balanced optimal strategy based on sequential gaming is proposed to optimize the excuting Makespan of Applications and improve the resource utilization efficiency of a Grid system. It can get an optimal solution after a limited stage gaming. Every stage game uses the balanced idea to divide the resource allocation process into two stages, namely activity distribution and processor allocation. Moreover, an extended strategy which aims to minimize cost is developed. By using the data from a practical grid system and application, the extensive simulations were conducted to investigate the performance of the strategy in a famous simulation test-bed. The experimental results indicate that, as compared with the Min-min strategy, this optimal stratcgy has the considerably low time complexity, the better excut- ing performance of applications and the higher resource utilization efficiency. The optimal strategy distributes activities to some ＇good＇ candidate resource sites, which leads to bigger impact of inaccurate estimation information on Makespan.