On implementing MPI-IO portably and with high performance
Proceedings of the sixth workshop on I/O in parallel and distributed systems
zFS " A Scalable Distributed File System Using Object Disks
MSS '03 Proceedings of the 20 th IEEE/11 th NASA Goddard Conference on Mass Storage Systems and Technologies (MSS'03)
The Panasas ActiveScale Storage Cluster: Delivering Scalable High Bandwidth Storage
Proceedings of the 2004 ACM/IEEE conference on Supercomputing
Benefits of high speed interconnects to cluster file systems: a case study with lustre
IPDPS'06 Proceedings of the 20th international conference on Parallel and distributed processing
Hi-index | 0.00 |
Parallel I/O needs keep pace with the demand of high performance computing applications. I/O bottleneck still is one of the key problems which restrict the development of high performance computer. Object-based storage combines the advantages of SAN and NAS, adopts an object-based interface, and has been a new orientation of massive storage. The trend as such makes it very promising to build a scalable I/O system for highperformance computing through implementation of object-based storage system based on high-performance interconnects. Lustre is a high performance distributed object parallel file system, is being widely studied and applied. Exploiting performance of Lustre depends on supports of high performance interconnects. In this paper, firstly we designed and implemented a customized high performance interconnect chip, HSNI (High performance Storage Network Interface), and introduced the characteristics of HSNI. Then we analyzed the communication features which support high performance of Lustre. Experimental results show that HSNI can greatly exert the performance of Lustre and has a very good scalability, so its very suitable for building the high performance object-based storage systems for high performance computing.