Showcasing NeoSapphire Models for HPC Applications and Debunking Common Myths Surrounding All-Flash Performance
AccelStor, the software-defined all-flash array provider, announces its attendance at SC15, the International Conference for High-Performance Computing, Networking, Storage, and Analysis. From November 16th to 19th, 2015 in Austin, Texas, AccelStor will be at booth 2209 in the Austin Convention Center showcasing its 1U model, NeoSapphire 3505, and 2U model, NeoSapphire 3413, and demonstrating HPC storage acceleration using all-flash arrays. In addition to its main exhibit, AccelStor’s Vice President of Research and Development, Dr. Weafon Tsao, is holding a speech session on Thursday, November 19th at the Exhibitor Forum to debunk common myths surrounding all-flash storage and introduce AccelStor’s software-defined approach to flash acceleration.
NeoSapphire All-Flash Array Series for High-Performance Computing
All-flash arrays represent the pinnacle of storage performance and at SC15 AccelStor is showcasing some of its highest performing models for accelerating HPC application performance. The NeoSapphire 3505 packs over 500K sustained IOPS for 4KB random writes into a rack-space saving 1U form factor. With the high-speed 56Gb/s InfiniBand FDR connectivity, the NeoSapphire 3505 has the throughput and response time to meet the demands of higher performance computing. AccelStor also has on hand the 2U NeoSapphire 3413, an over 600K sustained IOPS flash appliance with 13TB usable capacity for higher capacity flash storage needs over mainstream 10GbE network interfaces.
HPC Application Acceleration Live Demo
As compute and memory performance increases, storage performance increasingly becomes the bottleneck in modern HPC applications. At booth 2209, AccelStor is showing a live demonstration of a typical highly parallel, large-footprint HPC application connected to NeoSapphire 3505 all-flash storage.
Running a dozen of concurrent, heavy-loaded pathfinding applications using the well-known A* algorithm, AccelStor demonstrates how the sustained IOPS performance of its NeoSapphire 3505 all-flash array can accelerate this typical graph-searching problem commonly seen in HPC applications. It is not unusual for HPC applications to grow into large footprints, where the working set grows beyond the capacity of volatile memory stores during intensive computation. In these cases, excessive swap operations to persistent storage reduces computational efficiency of the HPC application. With a native high-speed InfiniBand FDR 56Gb/s network interface and a usable capacity of 5TB, NeoSapphire 3505 offers an affordable storage solution that greatly improves the efficiency of HPC applications while reducing total cost of ownership.
Sustained High IOPS Performance
At SC15, AccelStor is demonstrating NoeSapphire 3413, a 2U all-flash array which can provide the sustained random IOPS performance of over 600K IOPS on average in extended-hour stress tests. All NeoSapphire all-flash arrays need to pass extended-hour stress tests and in these tests the NeoSapphire 3413 has shown performance of 630K IOPS for 4K random writes over a 24-hour period. Real-world HPC applications need more than high peak performance. As flash appliances often provide high-speed, shared storage services to multiple front-end compute nodes, high sustained IOPS is the real driver in flash storage performance for I/O-intensive HPC applications.
Speech Session – Debunking Myths about All-Flash Performance
At the Exhibitor Forum, AccelStor’s Vice President of Research and Development, Dr. Weafon Tsao, is giving a presentation, “Building a High IOPS Flash Array: A Software-Defined Approach,” debunking a number of myths surrounding flash storage performance and giving insight into the technology behind AccelStor’s NeoSapphire all-flash arrays. Dr. Tsao’s speech will be given at 11:30am on Thursday. November 19th at room 12AB.