Supercomputer upgrade is under way

Mike Hawkins

The four racks of Cray Sonexion
First delivery. The four racks of Cray Sonexion storage delivered in February will provide an extra five petabytes of storage, enough space to store a billion copies of the complete works of Shakespeare.

On 19 February 2016, the Centre received the first pieces of equipment supplied as part of a significant upgrade of its supercomputers. The additional Cray Sonexion 2000 storage was delivered under a 36-million-dollar contract signed earlier this year with Cray Inc. The deal will see the current Cray XC30 supercomputers upgraded and expanded to Cray XC40 systems.

The upgrade involves adding a new cabinet to each supercomputer, taking it from 19 cabinets to 20, and swapping all the existing compute blades in each machine for new ones with the latest generation of Intel Xeon processors.

Another step in the upgrade took place at the end of March, when there was a short outage to install new nodes. A seven-week user test period is now under way, during which 25,000 cores will be available on one cluster for people to test codes at scale. This testing step is extremely important: we need to be sure that we can run our forecast codes on the upgraded machines and be sure of the results before we commit to the next stage in the process. In the next step, Cray will take each supercomputer out of service for a week to swap approximately 900 blades in each machine. When the upgrade is complete, the result will be two identical machines, each with more than 130,000 cores of the latest Intel Xeon ‘Broadwell’ processor – 55% more than a current cluster – and more than double the amount of memory. Cray will also supply a 32-node Cray XC40-AC system with the next generation of the Intel Xeon Phi processor.

This upgrade will help us to improve the quality of the service we provide to our Member and Co-operating States,” ECMWF Director of Research Erland Källén said. “It will enable us to develop high-resolution ensemble forecasts that improve the prediction of severe weather events in the medium range, up to about two weeks ahead. It will also make it possible to introduce improved data assimilation methods, allowing us to use more of the available Earth system observations, and to produce more detailed and better-quality atmospheric composition forecasts as well as high-quality climate datasets (reanalyses).”