Parallel I/O Optimizations Through Request Delegation for High-Performance Computing Systems.

摘要

Massively parallel applications often require periodic data checkpointing for program restart and post-run data analysis. For many high performance applications uncoordinated I/O is critical, particularly for the applications whose data is dynamically created or irregularly partitioned. In particular, parallel programs generating a large number of uncoordinated accesses on large scale systems often face serious I/O serializations introduced by lock contention and conflicts at file system layer. As these applications may not be able to utilize the I/O optimizations requiring synchronization, they pose a great challenge for parallel I/O architecture and software designs. Although high performance computing systems provide massive parallelism and computing power to fulfill the crucial requirements of the scientific applications, the I/O tasks of high-end applications do not scale. Pessimistic I/O strategies adopted from traditional file systems are inadequate for homogeneous parallel compute platforms. We propose an I/O software architecture to bridge the gap between scientific applications and parallel storage systems. A static file domain partitioning method is developed to align the I/O requests and produce a client-server mapping that minimizes the file lock acquisition costs and eliminates the lock contention. Our performance evaluations using production application I/O kernels demonstrate scalable performance and achieve high I/O bandwidths.