Multi-Tiered Storage Solutions – ClusterStor Data Services
TimeWednesday, 20 November 201910:30am - 11am
DescriptionStorage solutions for HPC continue to grow faster than the computational facilities as the amount of data (both used for analysis and generated by simulations) keep increasing at an exponential rate. To alleviate bandwidth and IOPS bottlenecks in the I/O path, many turns to tiering with flash-based solutions handling the most demanding workloads, HDD based capacity storage and tape or cloud storage systems acting as long-term data repositories. While this approach improved computational performance and solve the problem with bandwidth and IOPS related I/O, it creates new problems with data placement and expensive data movement.
Cray has been working on technologies and methodologies trying to solve these problems for a long time and by utilizing modern features in open source file systems such as Lustre, developed the means to optimized data placement, minimize data movement while maintaining a coherent view of the name space. We call this ClusterStor Data Services.