跨地域分布数据中心高成本效益的任务调度

Cost-Efficient Task Scheduling in Geo-distributed Datacenters

研究了跨地域分布数据中心云计算成本最小化问题,首先将其建模为一般分配问题,利用增广拉格朗日乘子法(augmented Lagrangian multiplier method,ALMM)得到最优的调度方案,然后设计了Adjusting算法以调整ALMM产生的结果,使之成为可行解.进一步设计了一种降序价值密度算法(decreased value density scheduling algorithm,DVDS),以解决ALMM的收敛速度相对较慢的问题.在线性定价和阶梯定价两种模型下的实验结果表明,当任务数量较少时,DVDS算法可以在拥有极小的时间开销下达到和ALMM同样小的调度成本;当任务数量增加时,DVDS产生的调度结果成本相比ALMM仅增加10%左右.

In this paper,we study the cost minimization for cloud users in geo-distributed cloud systems.By modeling it as a general assignment problem(GAP),we use the augmented Lagrangian multiplier method(ALMM)to obtain the optimal schedule solution.We additionally apply an Adjusting algorithm that adjusts the solution produced by ALMM to make it more feasible.We further use a decreased value density scheduling algorithm(DVDS)to speed up the convergence of ALMM.Experimental results show that DVDS algorithm can work out solutions in a much shorter period than ALMM does with costs similar to ALMM’s in the case of small task,and only 10%more in the case of large task.