Efficient Support for MPI-I/O Atomicity Based on Versioning

Abstract

Abstract:We consider the challenge of building data management systems that meet an important requirement of today's data-intensive HPC applications: to provide a high I/O throughput while supporting highly concurrent data accesses. In this context, many applications rely on MPI-IO and require atomic, non-contiguous I/O operations that concurrently access shared data. In most existing implementations the atomicity requirement is often implemented through locking-based schemes, which have proven inefficient, especially for non-contiguous I/O. We claim that using a versioning-enabled storage backend has the potential to avoid expensive synchronization as exhibited by locking-based schemes, which is much more efficient. We describe a prototype implementation on top of ROMIO along this idea, and report on promising experimental results with standard MPI-IO benchmarks specifically designed to evaluate the performance of non-contiguous, overlapped I/O accesses under MPI atomicity guarantees.

Towards a Storage Backend Optimized for Atomic MPI-I/O for Parallel Scientific Applications

“…The results have been reported in [15]. Our approach demonstrated excellent scalability under concurrency when compared to the Lustre-based approach.…”

“…We built the metadata corresponding to update made by non-contiguous writes such that only consistent snapshots that obey MPI atomicity are published. A detail implementation of our prototype is described in [15]. To leverage our storage back-end at the level of the MPI-I/O layer, we integrated our prototype into ROMIO [5], a popular MPI-I/O implementation.…”

Towards scalable array-oriented active storage

“…Pyramid is inspired by BlobSeer [11,17], a versioning-oriented storage system specifically optimized to sustain a high throughput under concurrency. BlobSeer focuses on unstructured Binary Large OBjects (BLOBs) that represent linear addressing spaces, whereas Pyramid builds on similar design principles to specifically address the case multi-dimensional data.…”

A Checkpoint of Research on Parallel I/O for High-Performance Computing

“…This approach may present poor performance under a high level of concurrency. Tran et al [101] propose a versioning approach to provide atomicity with better performance.…”