Chapter 3 Sockets
In this chapter, we will build upon the knowledge obtained so far to create an MPI subset capable of blocking and nonblocking point-to-point communication over Ethernet and other IP capable networks. Being about the slowest and most forgiving of all fabrics, Ethernet will allow us to go to the other extreme in the performance range and enjoy a little more freedom in the programming techniques. Again, this will not be a full MPI implementation, but it will come a step closer to the Real Thing.
3.1 Subset Definition
Let’s take the next reasonable step and make our new MPI subset a bit more flexible. Sockets are normally about 10–20 times slower than the shared memory as far as latency is concerned, thus we will be able to deal ...
Get Inside the Message Passing Interface now with the O’Reilly learning platform.
O’Reilly members experience books, live events, courses curated by job role, and more from O’Reilly and nearly 200 top publishers.
