Abstract
The COMPASS tokamak is expected to produce several gigabytes of data per shot in near future. A new storage system is needed to accommodate and access all the data.
It should be scalable, fault-tolerant, and parallel. It should not be based on proprietary solutions-to maintain independence from hardware and software manufacturers and preferably it should be built on inexpensive commodity hardware.
One of the promising distributed parallel fault-tolerant file systems, GlusterFS, was selected for testing. The aim of the work was to make initial tests of a particular small GlusterFS setup to confirm its aptitude for the COMPASS storage system.
Aggregated reading throughput from multiple NFS clients was one of the most important figures that were benchmarked, it scaled well with the number of clients, starting just above 60 MiB/s(1) for 1 client and going slightly over 300 MiB/s for 6 clients.