A Testing Engine for High-Performance and Cost-Effective Workflow Execution in the Cloud

Abstract

While pursuing high performance and cost effectiveness for directed acyclic graph (DAG)-structured scientific workflow executions in the cloud, it is critical to identify appropriate resource instances and their quantity. This paper presents a testing engine that employs a resource-selection heuristic, which statically analyzes the DAG structure to guide the selection of resource instances, how many and which ones. The testing engine combines the heuristic with two platformindependent DAG-scheduling policies, the Area-oriented DAGscheduling heuristic AO (AO) and the Locally-Optimal heuristic (L-OPT), to perform extensive validation assessments. The testing engine ensures the realism of these assessments by modeling the performance variability of the cloud platform using real traces. The testing engine also enables cost-effectiveness analysis that guides users to select a small set of instance candidates that provide performance-cost tradeoff. Our empirical results show that the pairing of the resource-selection heuristic with AO scheduling policy is a powerful method for cost-effective DAGstructured workflow execution in the cloud.

A2Cloud‐RF: A random forest based statistical framework to guide resource selection forhigh‐performancescientific computing on the cloud

“…Figure 36 shows the cost rating versus A2Cloud rating for the tested instances. Motivated by the Cloud performance‐cost trade‐off studied by Pallipuram et al, 63 the A2Cloud‐RF framework recommends a Cloud instance whose cost‐performance rating (( cost , A 2 Cloud )) is farthest (in Euclidean distance) from the lowest possible cost‐performance rating (( cost , A 2 Cloud ) = (1, 1)) and closest to the highest possible rating (( cost , A 2 Cloud ) = (4, 4)). Figure 36 elucidates that the c4.large and N1s2 instances have ( cost , A 2 Cloud ) ratings of (3, 4) and (4, 3), respectively.…”

Scheduling DAG-based workflows on single cloud instances: High-performance and cost effectiveness with a static scheduler

“…We use the weighted median to measure G ’s degree of concurrency because it is more robust to extreme values than many other statistical measures are and it is a less-biased statistical parameter (Brownrigg, 1984). Moreover, this metric captures the intuitive notion of degree of concurrency well for DAGs having similar dependency lengths but different patterns of concurrent tasks (Estrada et al, 2015; Pallipuram et al, 2015).…”

“…The homogeneity of the tasks within each DAG-workload and of the vCPUs within each instance allows us to estimate the runtime of the sample workflows for the five configurations in a simple simulation environment. Specifically, we use a cloud simulator previously calibrated in Estrada et al (2015) and Pallipuram et al (2015) to estimate lower-bound execution times and costs for the considered schedulers and resources.…”