Even though file system performance is not measured by the Linpack benchmark, which is the basis for the TOP500 list of the worlds fastest supercomputers, the efficiency of the parallel cluster file system nowadays is known to be one of the key aspects of cluster computing to avoid long application stalls on I/O wait. This factor is becoming even more relevant when considering the ever increasing size of data sets that need to be analyzed at higher levels of detail and the increasing size of application checkpoints due to the growing size of RAM on the compute nodes.
"Without a doubt, Intel has made a big leap in performance with the new 100Gbps OPA technology compared to previous interconnect generations. The fact that we didnt need to modify even a single line of the BeeGFS source code to deliver this new level of throughput, confirms that the BeeGFS internal design is really future-proof", commented Sven Breuner, CEO of ThinkParQ.
Marco Merkel, Global Sales Director of ThinkParQ, added: "We are already seeing a number of customer projects being implemented with BeeGFS over Omni-Path, where our solution partners were able to significantly scale-up the per-server throughput and thus needed fewer servers to exceed the customer expectations. Especially for all-flash systems with BeeGFS, which are getting very famous in the finance industry, this new technology will help to unleash the full potential of BeeGFS."
For people interested in trying out BeeGFS, ready-to-use binary packages for all major Linux distributions and also the BeeGFS source code are available free for download from the website http://www.beegfs.com . The available downloads already come with full support for Omni-Path. For those who would like to meet the team, the ISC High Performance Computing Conference in Frankfurt, Germany, starting on June 19, will be a good chance to get in touch.