Access

Dear colleagues,

This message is to update you on the time-line for withdrawing iVEC’s Epic system from service, at the end of the year, and to explain some of the alternatives that are and will be available at iVEC.

As many of you will be aware, the Epic service will close on 31st December 2014, having delivered more than 3 years of service to the iVEC research community, and having led to hundreds of publications, presentations and discoveries. The end date coincides with the end of the current merit-based compute-time allocations on Epic — that is, NCMAS, Geosciences, and the iVEC Partner allocations, plus the iVEC Director’s Share.

Epic is being replaced by our new, petascale service called Magnus, which — at the time of writing — is about to move into early-science mode, and which will be available through NCMAS, Geosciences, and iVEC Partner merit-allocation schemes from the beginning of 2015. Just like Epic, Magnus is a CPU-based system with a high-performance interconnect enabling large-scale parallel simulations and computations to be completed. Magnus, however, has more than three times as many compute cores as Epic, has more than ten times the compute capacity, and has five times the working-storage (that is, scratch) provision. We envisage that the majority of Epic-based research projects — which rely on multi-node parallel computing — will be able to migrate to Magnus with minimal porting effort and indeed many of you will already have had the chance to start this porting work on the Magnus Phase 1 system.

Alongside the many parallel applications that run on Epic, there are some smaller-scale computations and data-intensive processes undertaken. These may run on single nodes but exploit the opportunity to run many concurrent simulations (across multiple nodes) and perhaps employ within-a-node parallelisation techniques such as OpenMP or threads. These applications may not be well-placed to take advantage of the specialised architecture of Magnus, and may be better suited to two other services that will move in production in the near future.

The first of these is the iVEC NeCTAR Research Cloud node, which is a part of the wider NeCTAR Cloud [1]. The iVEC NeCTAR Research Cloud node is expected to be in full service in the last quarter of 2014. As well as contributing to the National Cloud Federation it will have a portion of the node targeted at data-intensive applications supporting iVEC-community activities. iVEC staff have evaluated the opportunities that the NeCTAR Research Cloud offers, and will be providing guidance and support to help researchers migrate onto the NeCTAR Research Cloud where appropriate.

The second service that may be of interest is called the Zeus cluster, which includes 30+ nodes in various configurations aimed at supporting remote-visualisation work, pre- and post-processing computations, and large shared-memory calculations (including the already-in-production UV2000 system, called Zythos). The Zeus cluster is, at the time of writing, in early-adopter mode. It is expected to be in production in the next month or so.

At the beginning, the Epic system represented a major increase in the computing capability that was available to the iVEC research community. It has been pleasing to see how well the community has embraced the opportunities it has provided. The Magnus system, NeCTAR research cloud node, and the Zeus cluster provide yet another significant increase in iVEC’s compute capability and I look forward to seeing how you exploit this capability going forward.

If you have any questions or concerns regarding the closure of Epic, please do not hesitate to contact the iVEC Helpdesk [2]. Otherwise, keep an eye on up-coming announcements, regarding the arrival of the new systems into full operation and the ways in which the Supercomputing Team will be helping people to migrate across to these systems.

Kind regards,

George Beckett.