Learning-based Dynamic Pinning of Parallelized Applications in Many-Core Systems

TL;DR

This paper introduces a learning-based, decentralized scheduling framework for dynamic thread placement in NUMA architectures, enhancing performance and adaptability of parallel applications in many-core systems.

Contribution

It presents a novel decentralized learning scheme for thread placement that easily incorporates multi-objective criteria and adapts to runtime performance changes.

Findings

Performance improvements over Linux scheduler, especially with limited resources.

Significant gains in irregular memory-access patterns.

Analytical guarantees on expected application performance.

Abstract

Motivated by the need for adaptive, secure and responsive scheduling in a great range of computing applications, including human-centered and time-critical applications, this paper proposes a scheduling framework that seamlessly adds resource-awareness to any parallel application. In particular, we introduce a learning-based framework for dynamic placement of parallel threads to Non-Uniform Memory Access (NUMA) architectures. Decisions are taken independently by each thread in a decentralized fashion that significantly reduces computational complexity. The advantage of the proposed learning scheme is the ability to easily incorporate any multi-objective criterion and easily adapt to performance variations during runtime. Under the multi-objective criterion of maximizing total completed instructions per second (i.e., both computational and memory-access instructions), we provide analytical guarantees with respect to the expected performance of the parallel application. We also compare the performance of the proposed scheme with the Linux operating system scheduler in an extensive set of applications, including both computationally and memory intensive ones. We have observed that performance improvement could be significant especially under limited availability of resources and under irregular memory-access patterns.