Oikonomou, Panagiotis and Kolomvatsos, Kostas and Loukopoulos, Thanasis}
24th Pan-Hellenic Conference on Informatics
Publication year: 2020

In the generic statement of the workflow scheduling problem we are given a set of tasks that have precedence constraints and a set of available machines whereby they can be assigned. The problem is to schedule the tasks to machines so that some target function, e.g., makespan, is optimized. The workflow itself is usually modeled by means of a DAG with nodes representing tasks and edges capturing precedence requirements. In the multiple workflow version of the problem a set of workflows must be scheduled concurrently to the available machines. Most research in the area focused on using a common task queue for all workflows, while task to machine assignment exhibited no restrictions, other than the ones imposed by performance optimization criteria. Nevertheless, strong privacy requirements might entail disjoint workflow to machine assignments. In this paper we tackle the resulting resource allocation problem when multiple workflows must be executed in a secluded manner. We investigate the performance of three heuristics that decide upon the splitting of available machines to the workflows. Experimental evaluation using common benchmark DAGs reveal different trade-offs for the proposed schemes.