Linux vs. lightweight multi-kernels for high performance computing: experiences at pre-exascale

ABSTRACTThe long standing consensus in the High-Performance Computing (HPC) Operating Systems (OS) community is that lightweight kernel (LWK) based OSes have the potential to outperform Linux at extreme scale. To explore if LWKs live up to their expectation we developed IHK/McKernel, a lightweight multi-kernel OS designed for HPC, and deployed it on two high-end supercomputers to compare its performance against Linux. Oakforest-PACS, an Intel Xeon Phi (x86) based supercomputer, runs a moderately tuned Linux distribution. Fugaku, the world's fastest supercomputer at the time of writing this paper, is based on Fujitsu's A64FX (aarch64) CPU that runs a highly tuned Linux environment. We discuss recent developments in our OS and provide a detailed description on the challenges of tuning Fugaku's Linux for high-end HPC. While in a moderately tuned environment McKernel significantly outperforms Linux (by up to approximately 2X), on Fugaku we observe an average of 4% speedup across all our experiments, with a few exceptions where the LWK outperforms Linux by up to 29%. As part of our evaluation we also disclose a full scale (158,976 compute nodes) noise profile of the Fugaku system.