Today Terascala announced that an HPC storage appliance managed by Terascala is simplifying Lustre file system management at Virginia Tech’s Advanced Research Computing (ARC) facility.
“DDN presents some key features of the SFA12KX such as partial rebuild and QoS for reads which are quite unique in HPC. We also discuss the latest performance results from Lustre benchmarking, particularly as they apply to awkward workloads such as seen in genomics or energy exploration, that require both streaming and transactional performance. In this talk, we present an evaluation of transactional performance and scalability in Lustre today, including random read and write performance, small file performance, and metadata performance as run on DDN SFA12KX systems.”
“The acceptance and adoption of ClusterStor since its market introduction is a testament to the experience and stability that comes with a solution provider such as Xyratex,” said Earl Joseph, IDC program vice president for HPC. “With the release of the CIFS/NFS Gateway, we expect further market acceptance given that the Gateway aims to open avenues to Windows and Linux environments that were previously unavailable for the Lustre file system.”
“HPC systems have kept growing in the last decades, producing vast amounts of data. Over time, we noticed how increasingly difficult it is to manage these data. Lustre as parallel filesystem is widely used in the HPC domain. It answers most of the requirements of the new generations of the HPC systems. Before Lustre 2.4, the metadata operation was managed by single metadata server and target. Such single operation became a kind of bottleneck. As a first step to help solving this problem Lustre multiple metadata server and targets (DNE) came out with Lustre 2.4. This talk will cover the first experience with Lustre DNE, showing some results and discussing the next step.”
In this video from the 2014 HPC Advisory Council Spain Conference, James Coomer from DDN presents: Adventures in Parallel Filesystems. “IT is currently in the middle of a rapid evolution in both hardware (flash-technolo- gies) and software (object-storage approaches). With respect to parallel filesystems these advances can potentially both speed up IO and better cope with the volume of data – all at lower cost. We report the results of a number of technology evaluations around parallel filesystems and provide some insights into how flash and object-stores will impact HPC in the coming years.”