By Anthony J. Lockwood
Dear Desktop Engineering Reader:
I live in a population-challenged area, and Fat Ernie’s Cable TV & Internet Co. went belly-up, forcing me to switch over to DSL. Now, DSL is good, reliable, and quick enough for most of what I do. But not all. Sending and receiving huge files is plagued by latency and propagation delays. My neighbor, who never had broadband, doesn’t see it because he doesn’t know what he has missed. A similar effect can happen to organizations that migrate to HPC clustering. They don’t see that their speedy new rig really is not optimized for the application performance they should have. It’s just better than before.
Recognizing this, the folks at Microway developed an HPC simulator they call IBSim. What IBSim does is it allows a network designer to simulate the performance of classes of applications running on an InfiniBand cluster, quantify what’s going on, and develop cost-effective provisioning of your HPC clusters for what you’re really doing at your outfit.
IBSim’s magic is that it attempts to capture the essential features of an InfiniBand network – we’re talking about the hardware functionality that can lead to compromised performance – so that you can compare different fabric topologies. By simulating the computation and communication of the simulation, IBSim can help lead you to answers to such questions as how long the MPI simulation of the physical process will take under various hardware scenarios. All of which means you can tune your HPC set up to your unique work, rather than using a “one size fits all” arrangement.
Now, Microway uses IBSim to work with you on getting the best from them. You may not be in the market for new HPC components today, but that doesn’t matter. Today’s Check It Out white paper called “InfiniBand MPI Network Design and Simulation Using IBSim” by Paul Howard describes IBSim, and Mr. Howard has done a terrific job with it. Anybody, especially you network designers, will benefit from reading this paper. In particular, if you ever wondered if you are getting the bang for buck, this paper will fill you with all sorts of ideas that may point you where you need to look and what you need to understand about optimizing your HPC set up for your applications workload.
So, hit the link over there and download this valuable paper. You have a tremendous investment in your HPC cluster. You’ve earned the right to have it optimized for the performance it could give you. With expert help from the folks at Microway, you just might realize your HPC cluster’s fullest potential.
Thanks, pal. — Lockwood
Anthony J. Lockwood
Editor at Large, Desktop Engineering