An in-band and out-of-band cooperative operation and maintenance management system for supercomputing clusters

Abstract

The present application relates to the field of supercomputer cluster operation and maintenance, in particular to an in-band and out-of-band collaborative operation and maintenance management system for supercomputer clusters, which comprises data acquisition, processing, collaborative decision-making and execution control modules; the data acquisition module allocates task queues using the minimum weight priority scheduling algorithm, matches tasks and devices using the consistent hashing algorithm, and acquires in-band and out-of-band parameter states; the data processing module maps data to the twin base through the digital twin algorithm, introduces spatiotemporal correlation anomaly detection, supercomputer operation and maintenance knowledge graph and entity linking algorithm, and forms a global data set unified in space, time and semantics; the collaborative decision-making module adjusts the in-band and out-of-band weights based on a dynamic weight model, constructs energy consumption prediction and fault risk assessment functions, and optimizes them through reinforcement learning; the execution control module generates inspection tasks according to risk classification, plans paths using an improved A* algorithm, optimizes air conditioner load and server power in combination with energy consumption prediction, and improves operation and maintenance efficiency and energy utilization efficiency.

Images