CSIRO removes HPC 'bottleneck' with storage upgrade

Australia's peak science body CSIRO is swapping out the scratch filesystem used across all of its high performance computing clusters to overcome a performance bottleneck created by its newest supercomputer.

The science and research organisation has signed a five-year deal with Dell-EMC valued at just over $3 million for the replacement filesystem used to store temporary or working data.

The new filesystem will be shared across all of the CSIRO's in-house supercomputers, including the Pearcey supercomputer, which is also undergoing an upgrade of its own, and the new $4 million Dell-built Bracewell supercomputer.

A spokesperson told iTnews that the organisation moved to replace the existing filesystem following the arrival of Bracewell last year.

This was because users had quickly become “able to saturate the IO bandwidth of the existing scratch filesystem” through their work to address “frontier science problems”.

“As our users became accustomed to the new capability of the Bracewell cluster, we anticipated that the IO performance of the filesystem would become a bottleneck restricting the performance of some of our users' codes,” the spokesperson said.

“This upgrade will remove that bottleneck.”

The new filesystem will be based on Dell PowerEdge R740 servers and include a platform with 2 petabytes of NVMe-based storage from Intel, as well as a BeeGFS parallel filesystem to provide resilience.

It will allow CSIRO to “harness the full potential of its systems” when conducting computational modelling, data analysis and machine learning.

Some of the organisation's current work includes addressing “de-novo genomic sequence alignment and medical imaging analysis”.

Dell are also supporting installation to CSIRO’s Canberra data centre. Earlier this year CSIRO handed Canberra Data Centre’s a $16 million deal to continue housing its corporate applications and HPC clusters.

Pearcey update

CSIRO is also in the process of updating its Pearcey cluster with new Dell servers “to provide additional compute power for [its] researchers”.

The hardware upgrade will consist of nine PowerEdge C6400 2RU Enclosures, containing 36 PowerEdge 768GB C6420 servers, which will “result in a heterogeneous architecture”.

The new capability – as well as the new filesystem – is expected to be used by Data61’s decision sciences team to continue their work around biomechanics and mining processing equipment design.