Sunday, August 7, 2011

QLogic – Enough with the Misleading Publicity!

If you don’t have the technology to win, at least try to keep your integrity… well… maybe it is too late now…

I was debating if to write this post or not, but at the end, I got tired of vendors (and I must say that most of them are not like that) that publish misleading information and decided to call it out. Making the long story short, I am going after a new system for some CFD codes, and I came across a new white paper from QLogic. In that white paper they made some claims that since InfiniBand was designed more than 10 years ago, it is no longer a fit for today’s technology, while the QLogic proprietary version of InfiniBand is the right way to go. You probably already know what my opinion on proprietary solutions is, but let us skip that issue for now.

In order to prove their claims, QLogic published the following figure, comparing their proprietary “PSM” InfiniBand software to the InfiniBand specification based, the verbs software interface:

The first thing that I did not like is the misleading view of the graph – QLogic decided to start the graph at rating 250 instead of 0 – obviously to make the gap looks bigger than it really is, but this is not the major story here. In this graph QLogic claims to provide a superior performance advantage with “PSM” compared to the InfiniBand verbs – achieving rating of 335.7 compared to rating of 280.6 (higher is better). I decided to look for references to confirm or to refute their claim. The first reference that you can check is of course the ANSYS FLUENT benchmarks web page - On that page, you can easily find the QLogic performance numbers. “Surprisingly” enough, on the same page you can also find some performance results that were published by SGI (using the same CPUs) over the InfiniBand standard verbs interface. And guess what is the SGI published rating result for 32 nodes / 384 cores? I will make it easy for you…  it is 335.8! It is not 20% lower than PSM… not 10% lower than PSM… it is actually higher than QLogic PSM!

So what did I learnt from this? First, this is yet another encouragement to my decision not to use proprietary solutions, second - InfiniBand by the spec is a great solution for the today’s clustering technologies, and third - I will never believe to any publication made by QLogic. It is amazing to me how come they decided to “avoid” such a clear reference…

No comments:

Post a Comment