Inductive-bias-driven Reinforcement Learning For Efficient Schedules in Heterogeneous Clusters

Subho Banerjee · Saurabh Jha · Zbigniew Kalbarczyk · Ravishankar Iyer

Keywords: [ Bayesian Methods ] [ Deep Reinforcement Learning ] [ Monte Carlo Methods ] [ Hardware and Systems ] [ Applications - Other ]

[ Abstract ]
Wed 15 Jul noon PDT — 12:45 p.m. PDT
Wed 15 Jul 11 p.m. PDT — 11:45 p.m. PDT


The problem of scheduling of workloads onto heterogeneous processors (e.g., CPUs, GPUs, FPGAs) is of fundamental importance in modern data centers. Current system schedulers rely on application/system-specific heuristics that have to be built on a case-by-case basis. Recent work has demonstrated ML techniques for automating the heuristic search by using black-box approaches which require significant training data and time, which make them challenging to use in practice. This paper presents Symphony, a scheduling framework that addresses the challenge in two ways: (i) a domain-driven Bayesian reinforcement learning (RL) model for scheduling, which inherently models the resource dependencies identified from the system architecture; and (ii) a sampling-based technique to compute the gradients of a Bayesian model without performing full probabilistic inference. Together, these techniques reduce both the amount of training data and the time required to produce scheduling policies that significantly outperform black-box approaches by up to 2.2×.

Chat is not available.