Develop domain-specific web portals for submitting and managing corresponding compute jobs on the HECToR National Supercomputing Facility (http://www.hector.ac.uk/) in order to reduce the current failure rates and lower the barrier of uptake to new user groups.
Large computational simulations or analysis are becoming essential tools for a wide variety of scientists who would not previously have needed to use high-end computational facilities for their research. Although there is a huge amount of computational time available on the HECToR National Supercomputer (hosted and administered by the University of Edinburgh) it is difficult for novice users to utilise the resources effectively due to a lack of both technical expertise and familiarity with the interface and job submission system used. This problem effects every user of HECToR as up to 30% of all jobs fail due to errors in the jobs' batch script; sometimes after waiting as long as 3 days to run. A browser-based interface can remove these hurdles by providing canned solutions and placing constraints on values. Although web applications are routinely used for a wide variety of tasks, submitting jobs to the UK National Supercomputing facility is not one of them. A web interface is easily accessible to all researchers---whether they have a background in HPC or not---and has the potential to open up access to these powerful computing tools to scientists who cannot currently make use of them. This, in turn, enables new science to be performed. A major reason for not creating a browser-based solution is the cost and time associated, as well as the problem that different groups of users may want different interfaces to work with. Some projects have reported 12 or even 24 person months of effort dedicated to a single portal development. The UK National e-Science Centre has developed a technology that can deliver web-portals cost effectively and in a fraction of the time (weeks rather than months) compared to manually developed portals.
This exemplar project will be demonstrated to the HECToR user base in order to generate collaborations. We will also demonstrate it to Research Councils and promote it to potential HECToR user groups on the HECToR website via on-line videos (examples can be seen here http://research.nesc.ac.uk/taxonomy/term/77). The dialogue with specific research groups around the UK will open up avenues for future collaboration and funding bids between University of Edinburgh and these other research institutions for the development of more domain-specific portals. The reputation of the University of Edinburgh as an innovative host of large-scale HPC facilities will be enhanced across the UK, strengthening our bids for future hosting contracts. We will be able to promote the University of Edinburgh as a centre of excellence in providing web-based access to large HPC facilities. This will allow us to foster collaborations and apply for funding with other HPC facilities world-wide that aim to have their own web-based interfaces.
The success of the project will be measured by the feedback on the web portals received from HECToR users and also the number of future collaborations that result from this project. The HECToR service provider (EPCC) will continue to monitor the number of failed jobs, and we aim for a significant reduction by web-enabling applications that are used in jobs that have the highest failure rates. The major risk associated with this project is that there will be little uptake of the web interface from current HECToR users due to unforeseen social elements. We will include users from the start in the design of the portals to prevent inertia from current users of the facility.