PLFS

TLDR

Checkpointing shared single files with small, unaligned writes causes severe performance degradation on file systems optimized for large, aligned writes. The authors develop PLFS, a virtual parallel log‑structured file system, to reconcile application checkpointing patterns with file system performance. PLFS remaps an application's preferred data layout into one optimized for the underlying file system. Benchmarks on PanFS, Lustre, and GPFS show that PLFS can reduce checkpoint time by up to tenfold without requiring application changes.

Abstract

Parallel applications running across thousands of processors must protect themselves from inevitable system failures. Many applications insulate themselves from failures by checkpointing. For many applications, checkpointing into a shared single file is most convenient. With such an approach, the size of writes are often small and not aligned with file system boundaries. Unfortunately for these applications, this preferred data layout results in pathologically poor performance from the underlying file system which is optimized for large, aligned writes to non-shared files. To address this fundamental mismatch, we have developed a virtual parallel log structured file system, PLFS. PLFS remaps an application's preferred data layout into one which is optimized for the underlying file system. Through testing on PanFS, Lustre, and GPFS, we have seen that this layer of indirection and reorganization can reduce checkpoint time by an order of magnitude for several important benchmarks and real applications without any application modification.