CHECKPOINT/RESTART UTILITY FOR PARALLEL PROGRAMS ON ANUPAM - SUPER COMPUTER

Abstract

With the advent of Parallel processors and super computing technology the concept of "fault-tolerant computing" has emerged as an important issue. There are many Checkpoint/Restart Utilities available for sequential processors. But for large scientific computations which will take many hours for their execution, the obvious option is going for a parallel machine or super computer for fast execution. There is no such Checkpoint/Restart Utility available for programs running on parallel machine-ANUPAM super computer Developed by BARC. This Dissertation Work is an attempt to build an efficient Checkpoint/Restart Utility for parallel programs running on ANUPAM super computer. This is implemented using the concept of kernel level checkpointing and ZCF checkpointing schemes.- Optimization Techniques are used to minimize the checkpointing cost at each node of the parallel processor. A new checkpointing algorithm is designed and implemented which can minimize the checkpointing cost along with maintaining a consistent global checkpoint from which the parallel application can be restarted when a fault occurs. The Checkpointing scheme proposed takes communication induced checkpoints along with local checkpoints that are taken depending on the change in state size of the process. Finding a consistent global checkpoint is also easy in this scheme, which is an open question in ZCF schemes. This is implemented on Linux-8 operating system with kernel version 2.4.18 which is the official operating system used on ANUPAM. The language in which it is developed is