Certification of workflows in a component-based cloud of high performance computing services

Abstract

The orchestration of high performance computing (HPC) services to build scientific applications is based on complex workflows. A challenging task consists of improving the reliability of such workflows, avoiding faulty behaviors that can lead to bad consequences in practice. This paper introduces a certifier component for certifying scientific workflows in a certification framework proposed for HPC Shelf, a cloud-based platform for HPC in which different kinds of users can design, deploy and execute scientific applications. This component is able to inspect the workflow description of a parallel computing system of HPC Shelf and check its consistency with respect to a number of safety and liveness properties specified by application designers and component developers.