Abstract:
Recent trend is of high performance computing, in which message rate is of main measure. This message rate depends on communication pattern. The continued growth in platform scale, combined with emerging application area, are pushing platform to support increasing message rates. Best case message throughput has grown in hardware generation due to growing clock rates and on patterns.
Now, with the growing scale of high performance computing, fault tolerance has become the major issue. In computing work, there might be a chance of fault which will result in a false data. There are various techniques available to handle faults. A method named message logging is used to this overhead. There are various other techniques that are used to overcome fault tolerance.
Here I used an approach similar to the approaches of fault tolerance that exist. This approach is slightly differing with other existing approaches. With this approach we can handle faults in the communication patterns. For this, we have to change the parallel code of the communication patterns a bit.
