STORAGE

Panasas, Inc., in a joint effort with Los Alamos National Laboratory (LANL), today announced that the Panasas ActiveScale Storage Cluster has achieved more than 10 gigabytes per second of sustained data bandwidth for highly parallel Linux cluster applications(1). Three additional bandwidth records were set for tightly coupled cluster applications using LANL's MPI-IO benchmark. Together, these tests represent the range of throughput demands for technical computing applications. The Panasas storage cluster is currently deployed at LANL to provide scalable I/O for a variety of supercomputing clusters used for simulation. "Together, Panasas and Los Alamos are showcasing the architectural breakthrough of the Panasas storage cluster," said Garth Gibson, Panasas founder and chief technology officer. "Only an object-based storage design satisfies the range of I/O requirements for Linux cluster applications." "High-performance computing environments require high-performance data access and data storage," said Karl-Heinz Winkler of Los Alamos' Computing, Communications and Networking Division. "We are using Panasas' scalable parallel file systems and high-bandwidth storage solutions to build Linux-based HPC clusters for our demanding applications." Performance records for Linux clusters with MPI-IO The Panasas system also achieved 7.5 gigabytes per second bandwidth for barrier-bracketed N-to-N MPI-IO reads and 5.6 gigabytes per second for N-to-N MPI-IO writes.(2) These I/O scenarios are common in large-scale cluster computing applications such as seismic processing and physics simulations. Finally, the Panasas storage cluster delivered 2.8 gigabytes per second for N-to-1 concurrent writes, a 3X improvement over existing concurrent write solutions. Concurrent writes significantly improve data manageability by allowing the entire Linux cluster to write to a single file. Checkpoint applications, for example, can create thousands of files per day (one file per cluster node per period of time) whereas concurrent writing of checkpoints reduces the number of files created by several orders of magnitude. The Panasas storage cluster is now the world record holder for both random I/O and data bandwidth. On October 20, Panasas announced world record random I/O performance results of 305,805 SFS operations per second within a single Panasas system. Performance partners Extreme Networks, Rackable Systems and RackSaver provided critical components to delivering these high performance results. The Extreme Networks' BlackDiamond(R) core Ethernet switch enabled maximum bandwidth between the Panasas storage cluster and the Linux cluster comprised of high performance Rackable Systems and RackSaver servers. Panasas is a member of the Extreme Networks' Go Purple program and is working with Extreme in several accounts.