Following the training of NGCM PhD students in Threaded Programming on the UK’s supercomputer ARCHER, the students have joined forces with the Computational Modelling Group and plan to embark on an ambitious project: to fully automise high performance computing.
The roadmap outlines the basic strategy ahead: given an equation, the user will use voice recognition as available through Siri and similar technologies to input that equation into the high performance computing compiler. This new HPC compiler software will (i) automatically write state-of-the art code, (ii) make use of parallel computing paradigms such as OpenMP and MPI as well as emerging and future accelerator technologies, and (iii) automatically compile it. Although speech recognition has rarely been used for parsing equations in the past, the team believe this is possible now.
A sub group of students is working on an extension to the system that will automatically submit the compiled code as a job to be run on accessible supercomputers. On completion of the compute job — which could be several hours or days later — the scientist or engineer who communicated the equation to the extended Siri interface will be notified through email, twitter or snapchat. Eventually, depending on the user preferences and if the corresponding hardware is available and credit card details have been provided, the coffee machine will be started or a pizza will be ordered to be available just-in-time when the computation completes, to help the scientist analysing and understanding the numbers that the code has produced.
The team hopes to release the first open source version on 1 April 2016.
Posted by Hans Fangohr