Stochastic Tail-Phase Optimization for Bag-of-Tasks Execution in Clouds

Abstract

Abstract-Elastic applications like bags of tasks benefit greatly from Infrastructure as a Service (IaaS) clouds that let users allocate compute resources on demand, charging based on reserved time intervals. Users, however, still need guidance for mapping their applications onto multiple IaaS offerings, both minimizing execution time and respecting budget limitations. For budgetcontrolled execution of bags of tasks, we built BaTS, a scheduler that estimates possible budget and makespan combinations using a tiny task sample, and then executes a bag within the user's budget constraints. Previous work has shown the efficacy of this approach. There remains, however, the risk of outlier tasks causing the execution to exceed the predicted makespan.In this work, we present a stochastic optimization of the tail phase for BaTS' execution. The main idea is to use the otherwise idling machines up until the end of their (already paidfor) allocation time. Using the task completion time information acquired during the execution, BaTS decides which tasks to replicate onto idle machines in the tail phase, reducing the makespan and improving the tolerance to outlier tasks. Our evaluation results show that this effect is robust w.r.t. the quality of runtime predictions and is the strongest with more expensive schedules in which many fast machines are available.

Fast (re-)configuration of mixed on-demand and spot instance pools for high-throughput computing

“…We analyze these aspects on several workloads that have been found representative for real-world bag-of-tasks applications [14].…”

“…In previous work, we have demonstrated the efficacy of our BaTS scheduler for large bags of tasks on multiple cloud environments [13,14]. Without any a-priori knowledge of task runtimes, BaTS uses a tiny sample of tasks for estimating the price-performance ratios of the available types of cloud instances, for the given bag.…”

“…The case might always occur that one or more outlier tasks, with exceptionally high completion time, might be scheduled only towards the end of the overall execution. In [14], we presented an optimization for the tail phase execution that reduces the severeness of this problem.…”

“…We have evaluated the quality of the Pareto sets identified by our genetic algorithm using our BaTS simulator [14] that we have extended by price history logs from Amazon's spot market. We compare our solutions to offline computations of the exact Pareto sets, both in terms of quality and of computational effort needed.…”

Exploring portfolio scheduling for long-term execution of scientific workloads in IaaS clouds

“…Marshall et al [25] and Wang et al [43] present resource provisioning policies for parallel jobs in order to balance monetary cost and job wait time. Other efforts focused on cost-efficient execution of applications such as Bags-ofTasks (BoTs) [2,28] and scientific workflows [21,22]. To gain deep insight into the performance of scheduling policies, our previous work [41] studies the interplay between provisioning and allocation policies through real experiments.…”