Student Name : Manav Vasavada
Project Title: BLCR extensions for incremental checkpointing
Course : CSC 548 - Parallel Systems
Email : mmvasava@ncsu.edu

This project is about BLCR extensions for incremental checkpointing. Incremental checkpointing is an optimization for checkpoint restart mechanism to checkpoint only the modified area in the process address space. In a simple checkpoint, entire process is checkpointed at each checkpoint. However, some areas in the process which do not get modified frequently need not be checkpointed at each stage. We can mark all the modified areas and only checkpoint them. Doing so can save us disk space and time since the entire process space need not be dumped to the file. At restart file all the files including the first full checkpoint are overlapped one over the other with the most recent incremental checkpoint on top and the process image is reconstructed. Care should be taken about changes in the VM area over time. The restart process will have a lower performance then the simple restart process. The aim of this project is to compare two different approaches write bit and dirty bit approach and analyze their performances.

References:
"Hybrid Full/Incremental Checkpoint/Restart for MPI Jobs in HPC Environments" Chao Wang, Frank Mueller, Christian Engelmann, Stephen L. Scott
"Incremental Checkpointing for Grids", John Mehnert-Spahn, Eugen Feller, Michael Schöttner, Linux Symposium, Montreal, 2009

For the first report click here