Parallel File Systems
NFS is widely used in clusters and workstation environments, but the performance is restricted by the use of only one system operating as server. The parallel NFS, also referred to as pNFS, will overcome this restriction by providing scalable parallel access to files distributed among multiple servers. For the success of pNFS there are two requirements:
- Implementation of the NFS v4.1 client in the linux kernel
- Availability of pNFS server components, particularly open source solutions
After a first evaluation of pNFS proving the functionality was done, further testing will continue in order to study the feasibility of providing parallel NFS access on the Jülich data access server JUDAC to the data stored in the Jülich Supercomputing Centre distributed storage system dCache.
Jülich Supercomputing Centre co-ordinates the european project DEEP-ER, which evolves the architecture developed in the predecessor DEEP project (a HPC Cluster and a tightly connected HPC Booster built of many-core processors) to address two significant Exascale computing challenges: highly scalable and efficient parallel I/O and system resiliency.
For doing so the architecture makes use of Fraunhofer’s parallel file system BeeGFS (formerly known as FhGFS), with its distributed metadata architecture as a basis for parallel I/O. A prototype is being built at Jülich Supercomputing Centre and the I/O performance results will be evaluated and compared with results on actual production systems like Jülich supercomputers' file system JUST, which makes use of IBM's GPFS.
Contact: Cristina Manzano firstname.lastname@example.org