Chemistry’s New Cluster

By Penny Carmichael, on 1 March 2012

Whilst such titles are normally followed by the exploits of inorganic chemists, 2011-12 sees the dawn of a new computing facility within the department. The new supercomputer, known as the IB-Server is the result of a 7 research group collaboration with resident computing officer Jörg Saßmannshausen. The project, begun in 2010, will allow group members sole access to 420 cores with fast ‘Infiniband’ connections, dramatically shortening calculation times and enlarging the molecules and systems that can be investigated.

IB-Server joins the 5 pre-existing chemistry computer clusters in calculation for a wide range of projects. UCL has a long and excellent research record in computational chemistry, from crystal structure prediction (Professor Sally Price), to the modelling of selectively oxidative catalytic surfaces (Professor Nora de Leeuw). The impacts in terms of theoretical understanding, programming and contributions to on-going experimental investigations have been significant. Countless other computational researchers will benefit from IB-Server but the machine should also have a wider impact. As Dr Saßmannshausen explains, computational chemistry is a useful tool at the disposal of the whole department. Not only can computational screening highlight areas for further experimental research, but modelling allows insight into sometimes ‘unintelligible spectroscopic data’.

Jorg cooks up some computer spaghetti

The cluster itself comprises of 40 boxes each housing 4 nodes, with each node having 4 dual core ‘Intel L5630’ processors. The key feature though is a low latency ‘Infiniband’ connection allowing swift transfer of data between nodes. The effect of the connection is exemplified by the recent benchmark test carried out using the CP2K code; on the gigabit network a standard calculation took two and a half days, whereas the new IB-Server machine reduces the calculation time to less than 12 hours. Benchmark test results also placed UCL chemistry computers above ‘HECToR’, the UK’s leading academic computational research facility. Practically, the reduction in calculation time will greatly increase throughput for parallelised codes such as ‘VASP’, ‘CASTEP’ and ‘Quantum ESPRESSO’ (all implementations of Density Functional Theory) and will allow much more challenging systems to be studied.

As use of the departmental clusters increases, Jörg’s role in keeping the machines functioning correctly will be supplemented by consolidating the data backup system and providing areas for the department’s keen code writers to store programming projects. One single entry point for the UCL computational chemistry community to access programs and support for new users are also being considered. In the meantime, chemists needing a chemist’s advice on programs, implementation and hardware have expert help available.

- Jane Yates