Interference-driven resource management for GPU-based heterogeneous clusters

摘要

Systems and methods are disclosed that share coprocessor resources between two or more applications in a computing cluster using a job selector to receive jobs from a job queue; a node selector coupled to the job selector; an off line profiler with an interference prediction model; a coprocessor dynamic interference detection module; and a coprocessor interference response module.

主权项

1. A system to share coprocessor resources between two or more applications in a computing cluster, comprising:
an offline profiler with an interference prediction model; a job selector to receive jobs from a job queue; a node selector coupled to the job selector; a coprocessor dynamic interference detection module; and a coprocessor interference response modulewherein the node selector decides whether a performance degradation is acceptable, wherein Pi and Pi′ are lowest kernel frequency of job Ji without and with interference, respectively, wherein the node selector considers a first interference criteria that caps the slowdown toa predetermined threshold T, that is, max Pi/Pi′≦T, and wherein the node selector considers a second criteria that allows job JN nodes with JN, usingmax⁢{1Pi′,1PN′}≤K⁡(1Pi+1PN).