The GekkoFS file system jointly developed by the Storage Systems for Extreme Computing team from the Barcelona Supercomputing (BSC) and the Efficient Computing and Storage team from Johannes Gutenberg-Universität Mainz (JGU) has taken the number 4 spot in the IO500’s ‘10-Node Challenge’.
GekkoFS is a file system capable of aggregating the local I/O capacity and performance of an HPC cluster compute nodes to create an ephemeral high-performance storage space that can be accessed by an application in a distributed manner. This storage space allows HPC applications and simulations to run in isolation from each other with regards to I/O, which reduces interferences and improves performance.
GekkoFS is being developed in partnership with as part of the Horizon 2020 NEXTGenIO project and Germany’s SPPEXA programme.
The IO500’s ‘10-Node Challenge’ list is a global ranking that executes multiple concurrent processes in 10 compute nodes to benchmark the I/O performance of a HPC storage system. GekkoFS’ score of 125 ranks it fourth in IO500’s 10-Node Challenge List and ninth in IO500’s Full List, with an average 21.41 GiB/s of bandwidth and an average of 728,680 operations per second.
GekkoFS’ IO500 benchmark was run on the 34 compute nodes of the NEXTGenIO prototype cluster. NEXTGenIO, is an R&D project involving EPCC, Intel, Fujitsu, Arm, ECMWF, TUD, Arctur, and BSC which wasbeen granted €8 million in funding by the European Commission to prototype an I/O-specialized HPC cluster based on Intel® Optane™ DC persistent memory technology. Each of the prototype’s 34 compute nodes is equipped with two second-generation Intel® Xeon® Scalable processors and 3TB of Intel® Optane™ DC persistent memory, thus providing approximately 102TB of persistent I/O capacity to HPC applications.
Within the project, BSC collaborated with JGU in the design and development of GekkoFS, a file system capable of aggregating the local I/O capacity and performance of each compute node to produce a high-performance storage space that can be accessed in a distributed manner. This storage space allows HPC applications and simulations to run in isolation from each other with regards to I/O, which reduces interferences and improves performance.