Issues on High Performance Computing in India

A symposium on HPC Applications was held in IIT Kanpur on 12-14 March 2012.  It had around 100 participants with speakers from various disciplines.   We had a panel discussion on future of HPC in India.

Minutes of the Panel discussion of the Symposium on HPC Applications are given below:

Panelists: Rajat Moona (CDAC, Moderator), P. B. Sunil Kumar (IITM), Amalendu Chandra (IITK), Sanjay Mittal (IITK), K. G. Ayappa (IISc), M. K. Verma (IITK)

Many important points arose during the meeting by the panelists and participants.  Main points are listed below:

1. Manpower:  We do not have enough number of trained manpower for maintaining systems, writing programs, and HPC research.  The students are not sufficiently exposed to advances in computational tools.   Only formal course in this field in India is M. Tech. in "Computational Science"  offered by  IISc.  More institutes in the country should run such programs.

IIT Madras is offering a course on HPC with guest lectures from  industry. It was suggested that we should offer such a course  every semester and it should be accessible to students across  the country.  We could use video-conferencing for running this course across various institutions. NKN  is already providing a platform for such activities.

Coding is not rewarded sufficiently.  A science or engineering graduate is evaluated on the science content.  Also, the students do not see coding ability as an employability criteria. May be this  is not advertised sufficiently.  It was pointed out by  Nvidia that they are employing programmers.

2.  Big cluster vs. Small cluster:  Local HPC clusters are important for learning for the students and faculty. It is also important to have such facilities for training and code testing.   However small clusters are difficult to maintain.  Providing cooling and  power and  system maintenance is quite demanding and needs trained man power.  This is particularly difficult in smaller  universities where infrastructure is not up to the mark.   Therefore, we should probably have few  very large systems and   number of large clusters, which should be established at places where such expertise is available. This can networked via fast link (NKN ?) and made available to smaller universities. Maintainability is a big issue, which needs to be circumvented with strong industry collaboration.  The salary structure in the academia makes it difficult to create a professional supercomputing system, so creative methods needs to be worked out.

A large number of scientists and engineers should have access to  these machines.   A transparent accounting system could be set up.  A job scheduler that balances the load across these multiple institutions could be set up.   It may be desirable to follow western models for allocation of resources to the research groups to maintain accountability.

3. Education and collaboration forum:  We need active forums for dissemination of information to researchers, specially students.   Such a forum will also provide exposure to students active in HPC.  In addition  this will foster collaborations among research groups.  Possibly CDAC could take a lead in setting up and maintaining this forum.

There is a need for code and compute-resource sharing.  Forums should be created for such efforts.  A balance of Independent code development and code sharing should be kept in mind. We could  identify problem areas where codes can be developed with groups involving both computer scientists, programmers, possibly hardware developers along with scientists working on a specific problem. These could be India specific problems or problems of broader scientific interest.