Project Specification: Fault Tolerant Matlab*P 
Nisheeth Shrivastava
Rachit Chawla

---------------------------------------------------------------

Matlab*P is is a distributed library for doing scientific computations. 
The Matlab*P provides an interactive and easy way in which user can 
manipulate and do intensive computations on large data sets residing on 
parallel machines. These often take hours to complete which makes it 
susceptible to network or node failures.

In our project, we are going to develop a user-transparent fault tolerant 
library integrated into Matlab*P which successfully recovers a slave 
process from its failure. We are going to use a checkpoint based rollback 
recovery algorithm. A checkpoint will be used to recorded the execution 
state of slave process periodically. In case of a failure, the failed 
process will be restarted from the last checkpoint saved to the stable 
storage so that it does not need to do all the computations done so far. 

The aim is that the user need not write anything in the program to take 
care of checkpoint or recovery in case of any failures. The underlying 
protocol would do it without any explicit interaction from the user. 
Another objective is to minimize the changes required in the Matlab*P 
interface provided to the user.