End-to-end I/O Monitoring on Leading Supercomputers

This paper offers a solution to overcome the complexities of production system I/O performance monitoring. We present Beacon, an end-to-end resource monitoring and diagnosis for 40960-node Sunway TaihuLight supercomputer, currently fourth-ranked supercomputer in world. Beacon simultaneously collects correlates tracing/profiling data from all compute nodes, forwarding storage metadata servers. With mechanisms such as aggressive online offline trace compression distributed caching/storage, it delivers scalable, low-overhead, sustainable under use. Beacon’s deployment on more than three years, we demonstrate effectiveness with real-world use cases issue identification diagnosis. It has already successfully helped center administrators identify obscure design or configuration flaws, anomaly occurrences, interference, under- over-provisioning problems. Several exposed problems have been fixed, others being addressed. Encouraged by success monitoring, extend monitor interconnection networks, which is another contention point supercomputers. In addition, generality extending other Both codes part collected are released. 1