PPR-RM: Performance-to-Power Ratio, Reliability and Makespan — aware scientific workflow scheduling based on a coalitional game in the cloud

Abstract

A distributed scientific workflow scheduling scheme is proposed for maximized reliability and performance-to-power ratio (PPR) under an inevitable end-to-end delay (EED), i.e., makespan. The proposed algorithm is studied in a heterogeneous cloud computing infrastructure, where processing node and transmission link failures are unavoidable. Cloud nodes share the scheduling decision and the stored related data with their neighbors to maintain scalability and robustness, which are primarily significant for large-scale distributed schemes. This workflow scheduling optimization strategy (PPR-RM) considers the maximum reliability and PPR under EED constraints in a three-step procedure. In the first step, an optimal mapping algorithm is adopted to minimize the EED of the critical path’s tasks using the iterative critical path search and Layer-based priority placing techniques (CPP). The second step samples the node’s utilization levels with distinguished PPR, calculated as the number of transactions performed in a specific time divided by the average active power consumed. This step trains processing nodes at different utilization levels to guarantee those nodes operate at the most power-efficient utilization levels, i.e., the highest PPR levels. Further, this step tremendously decreases energy consumption without sacrificing EED and reliability. The final step reallocates the tasks on non-critical paths for maximum performance reliability. Our extensive results demonstrated that PPR-RM acquired considerably higher reliability values under the minimized EED and energy constraints than some competitor workflow mapping algorithms.