成本约束的网格工作流时间优化方法

Time Optimization Heuristics for Scheduling Budget-Constrained Grid Workflows

针对成本约束有向无环图DAG(directed acyclic graph)表示的网格工作流完工时间最小化问题,提出两个基于优先级规则的迭代启发算法.算法利用并行活动特征定义正向分层和逆向分层两个概念,将其分别引入最大收益规则MP(maximum profit),得到正分层最大收益规则MPTL(maximum profit with top level)和逆分层最大收益规则MPBL(maximum profit with bottom level).两规则每次迭代尽量以完工时间的最小增加换取总费用的最大降低,逐步将分层初始解构造为满足成本约束的可行解.模拟结果表明,两规则在获得较少迭代次数和运行时间的同时,能显著改进MP规则的平均性能,且MPBL优于MPTL.

Workflow scheduling which guarantees anticipated QoS （quality of service） is a fundamental and complex problem in grids. In this paper, the budget-constrained scheduling of workflows represented by DAG （directed acyclic graph） with the objective of time optimization is considered. In general, the optimization problem is NP-hard. Two new iterative heuristics based on priority rules are proposed. According to the property of parallel activities in DAG, two concepts called TL （top level） and BL （bottom level） are defined. By incorporating them with priority rule MP （maximum profit） respectively, two priority rules, i.e. MPTL （maximum profit with top level） and MPBL （maximum profit with bottom level） are designed. They are implemented in iterative heuristics to generate iteratively feasible solutions from leveling-based initial solutions which need bigger total costs. In each step, MPTL and MPBL manage to take into account the maximum decrease in the total cost but the minimum increase of workflow duration. Computational experiments show that MPTL and MPBL can considerably improve the average performance of MP within a few iterations and a little computation time. Moreover, MPBL outperforms MPTL. As well, the impact of budget constraints and the number of Web services on the two heuristics are analyzed.