Plenary Session on Affordable HPC
In several institutes and universities, there is one person or a few persons who need access to HPC facilities, but cannot afford to set up one themselves. Access to big systems in a few places, as envisaged under NKN, is a good step forward, but is likely to run into the bottleneck of heavy usage within these institutes.
It is also not feasible for most people to set up a HPC facility by themselves spending all the time ensuring power and cooling are taken care of. Our interest is in using HPC systems for computation, and if this goal can be achieved only by sacrificing a lot of time spent on running the facility, then most will simply look for a different problem to solve, a problem that does not demand such compute power.
There are many potential users of HPC facilities scattered across India. If they are given access to what we can call “Hassle Free HPC” or "Affordable HPC", then they will be willing to use these systems.
The hassle in using a HPC facility is four-fold.
First, there is the problem of procurement. Given the complexity of the machines, it is not easy to decide what to buy. This problem can be solved by an institution, which has the required expertise and is already applying this domain knowledge in several places.
Second, there is the hassle of running the hardware: power, backup generators, and cooling. This is probably the biggest hurdle, the one at which most would collapse.
Third, there is the problem of maintenance. Again, the required expertise is available with certain institutions, but there has to be a proper way out to service a user community scattered across India. If the systems are bundled into a few locations, this problem can be managed easily. One system administrator can handle several systems sitting in the location.
Fourth, using these systems demands knowledge of parallel programming.
So the plenary session on “Affordable HPC” will bring in picture various new models from intelligent HPC user community paving a right way for usage of HPC resources for various HPC users in varied scientific domains.