ICTS high performance computing underpins UCT research

It's been two-and-a-half years since Information and Communication Technology Services (ICTS) at UCT first ventured into the realm of high performance computing (HPC).

In 2009, the executive director of ICT at UCT, Sakkie Janse van Rensburg, challenged ICTS to provide an affordable and accessible HPC solution, encompassing hardware, software and user support for the university's research community. Two senior engineers from ICTS's Computing Platforms team, Timothy Carr and Andrew Lewis, were chosen to drive the project.

With limited hardware and no prior experience in HPC they were fortunate to join forces with the South African National Grid (SAGrid) initiative, co-ordinated by Dr Bruce Becker at the Meraka Institute of the Council for Scientific and Industrial Research. This enabled the ICTS's fledgling HPC team to attend several international training events in Europe, build contacts with other South African universities and research groups, and gain access to invaluable support and experience.

In addition to running the ICTS-dedicated research cluster, the team plays a key role in supporting the SAGrid project, allowing UCT researchers to gain access to additional resources both in South Africa and internationally via the European Grid Initiative (EGI).

HPC progress at UCT was slow initially, but after attending a European Union-funded application porting school in Sicily, the team gained sufficient confidence to deploy a local cluster dedicated to UCT research. They also began to convert scientific applications from standalone desktop to HPC format.

Since then there has been a rapid uptake of users and a steady demand for new applications. Over 20 scientific software programs have already been ported to HPC format and the team are currently working on additional programs.

The team currently supports users from nine university departments, and recently surpassed 10 years' worth of computing time - the equivalent of one computer working constantly for 10 years.

Number of UCT research jobs submitted since the ICTS HPC cluster went live in March 2010.

Local UCT researchers account for over 80 000 computing hours, while the remainder of the HPC infrastructure is dedicated to international collaborative projects. This is largely due to the fact that it is possible to submit multiple jobs simultaneously to the HPC cluster.

It is worth bearing in mind that jobs can run for a month or longer, so ICTS ensures a stable and highly reliable service by making sure that the ICTS data centres are equipped with an uninterruptible power supply (UPS), in-line diesel generators, environmental control and disaster recovery systems. Once a job is running on the cluster, the researcher can carry on with other work on his or her personal computer without having to worry about processor utilisation or power failures.

Along with providing application support, the team also makes sure that ICTS has sufficient hardware to support the HPC requirements, and regularly adds to the number of CPUs (central processing units) that are available.

Later this year, ICTS also hopes to provide support for applications that require extreme computing power, and will be looking to expand its disk-storage pool into the multi-terabyte range.

The goal is to provide a low-cost, sustainable HPC architecture and to increase and retain HPC skills inside the university. Ultimately this will make ICTS HPC a more affordable and attractive model with which to do research.

Jobs submitted by UCT and international researchers to the ICTS Grid Cluster.

Find out more about the team by reading the ICTS HPC blog.