Increasing waiting time satisfaction in parallel job scheduling via a flexible MILP approach

Scheduling of jobs in parallel computing is crucial to efficiently use shared resources, while attaining user satisfaction. In this paper, we evaluate how mixed-integer linear programming (MILP) can be applied for the online parallel job scheduling problem (which is well-known to be an NP-complete problem). Therefore, we introduce the idea of planning horizons for parallel job scheduling, and provide a MILP formulation of the targeted scheduling problem. Due to the linear fashion of possible MILP objective functions, the proposed scheduling algorithm is flexible towards different optimization goals. We make use of data collected in a user-based study and workload traces from two real production systems, and demonstrate that our approach suffices to increase users' waiting time satisfaction. Additionally, we show that our MILP formulation outperforms the EASY scheduling technique with conservative backfilling, when neglecting the online character of job submissions.