Description
STFC Daresbury Laboratory – Long term plan with HPC for NP.
C. Unsworth and M. Labiche
Contact: Carl.unsworth@stfc.ac.uk; marc.labiche@stfc.ac.uk
High performance computing resources for the nuclear physics community.
Modern theoretical and experimental nuclear physics is inextricably linked to computing due to the complexity and volume of data involved. It can make economic sense to provide centralised community-owned high performance computing resources for some of the most computationally intensive applications. Some of these applications are highlighted below.
Monte Carlo Simulations
Monte Carlo simulation of radiation transport is used for development of new detector systems, optimisation of existing detector systems for particular experiments, and to aid in the interpretation of experimental results. MC simulations rely on the generation of pseudo-random numbers which is typically a CPU bound task. A centralised compute server with very high CPU power would be a valuble resource for such applications. MC simulations are intrinsically parallelisable and so multi core systems can be exploited but this sometimes requires significant work from the user to implement so such a system would ideally be chosen for per-core CPU performance.
Evaluation of Theoretical Models
Complex theoretical models can be CPU or memory bound depending on the specific case. A centralised computing resource with high total and per-core CPU performance together with very large RAM could be valuble for the theory community.
Data storage and backup
A centralised data storage and backup facility would help the nuclear physics community secure the of large data sets collected at great expense from labs around the world. In addition it would help scientists meet their responsibilities under open access data sharing agreements. Such a system would need a very large array of redundant hard disks for storage but would only have basic requirements for CPU and computing power.
Data analysis
Adding a powerful multi core CPU and larger memory to the data storage and backup system mentioned above would allow users to log in remotely and execute data analysis codes directly on the server. Such a system would reduce the need for time consuming transfers of large data between multiple locations and allow sorting to be completed and/or iterated more quickly. Such systems have been used to great effect in the high energy physics community and could be equally beneficial in nuclear physics but whether such a system would be economical depends on the particulars of a case e.g. data volumes, number of users, number of institutions, complexity of analysis procedures, etc.