Optimal scheduling of parallelizable jobs in multiserver systems

Establish optimal scheduling policies that minimize mean response time for parallelizable jobs in multiserver systems, where jobs can utilize multiple servers subject to parallelizability limits and arrive over time.

Background

The paper studies workloads composed of multiple classes of parallelizable jobs, each with a class-specific parallelizability limit and job size distribution, and seeks policies that minimize mean response time on k identical servers.

While this work derives asymptotically optimal policies in specific scaling regimes (LPF in Sub-Halfin-Whitt and SERPT in Super-NDS) and provides additional insights, the general problem of devising optimal scheduling policies for parallelizable jobs in multiserver systems is identified as open at the outset.