Lustre on Intel Xeon Phi delivered 10x the bandwidth of NFS as reported in the 2014 Lustre User Group (LUG) presentation “Running Native Lustre* Client inside Intel® Xeon Phi™ coprocessor” by Dmitry Eremin, Zhiqi Tao and Gabriele Paciucci of Intel Corporation. Network file systems are essential to the current generation of Knights Corner Intel Xeon Phi coprocessors because the native file system resides in the coprocessors RAM. Yes, saving a file on an Intel Xeon Phi coprocessor reduces the available memory on the device. The only way to avoid this issue is to use a network file system, which is why the Lustre 10x greater bandwidth is so important.
Lustre delivers 10x the performance of NFS on Intel Xeon Phi
The Lustre file system is a POSIX compliant, open source, parallel file system that supports the requirements of leadership class HPC and Enterprise environments. It looks and acts like any other filesystem yet scales to thousands of clients, petabytes of storage, and has demonstrated over a terabyte per second of sustained I/O bandwidth. Over 60% of the TOP100 supercomputers run Lustre.
For more information about the 10x result and the Intel Xeon Phi configuration, see the OpenSFS.org slides “Running Native Lustre* Client inside Intel® Xeon Phi™ coprocessor” and/or the following video from LUG 2014
For more information about Lustre on many systems see:
- http://opensfs.org/
- “Architecting a High Performance Lustre Storage Solution” that disususs Intel enhancements to Lustre.
You can also contact Intel directly about Intel Enhanced Lustre as illustrated by the graphic below, or learn more in the following video.
Intel enhancements to Lustre
Note that Lustre can exploit SSD storage quite nicely in a general-purpose CPU-based cluster environment
Luster SSD 16-node IOR performance
Lustre 128-node SSD IOR performance using 1 MB files
Leave a Reply