It’s been just under a year since the project team were given the go-ahead to start work on building this service. Those of you who attended the requirements-gathering workshops last summer may recall that we intend to launch the service in late Spring/early Summer of this year.
Although the project team is based in NUIT, we have invited input and participation from researchers at every stage of the project: gathering requirements, system architecture, staff recruitment and tender evaluation. More details will be posted in the coming week or so, but here’s a short summary of the projects progress to date.
People
After a couple of rounds of recruitment we are pleased to be welcoming a new HPC research computing analyst on board. This person will sit at the interface between the technology and the researchers whop use it, providing training, guidance and some coding support.
We are working with a recruitment agency in order to source a systems administrator to oversee the day to day management of the system itself.
The System
Summer’s requirements workshops and questionnaires, were followed up by a day-long solution architecture workshop, where a group of researchers, IT staff and some external HPC expertise evaluated and condensed the various technical requirements into a proposed solution that would serve a wide variety of potential users.
The result was a system based on the most recent generation of Intel Broadwell CPUs, connected via high-speed interconnect, in a topology that allowed for an island of nodes that would favour tightly-coupled parallel jobs. The system design also included groups of medium-, high-, and extra-high RAM nodes, featuring 0.25, .5 and 1.5 TB RAM per node. The 1.5 TB nodes feature 4 CPUs.
Omissions
There are several ways in which our envisaged service falls short of all of the requests for features that we received.Part way through the requiremnts-gathering phase of the project, something called Brexit weighed in. The tender did ask for indicative quotes for GPU nodes, and we have yet to make a final decision on that. We did not tender for any Intel Xeon Phi nodes, or nodes running Microsoft Windows. We are looking into the possibility of facilitating access to Intel Xeon Phi nodes at the Hartree facility, via the N8 consortium.