cluster

cluster / computing engine / high performance computing / hpc — View Comments
13
Feb 09

An Upfront CAE "Computing Engine" for the Common Man

The beauty of the hardware vendors making strides in technology is a great thing for all. It simply drives the market to a point where really powerful computing power is available to the common man, not just the elite early adopters.
It was only a few years ago that the idea of parallel computing was really only available to a select few. Parallel computing, of course, is the ability to leverage multiple machines connected together via a VERY high speed network connection for the purposes of doing some sort of computation FAST. Really fast.
Those of us in the Upfront CFD world or anyone in the CAE world for that matter, can always benefit from a bit more horsepower. Whether it is to help solve a really large model that can’t be solved on a standard workstation due to its size OR to reduce the amount of time a monster model takes to solve OR to speed up the process of your multiple what-if scenarios. In other words, all in an effort to SAVE TIME.

With Microsoft jumping into this world with Server 2008 HPC Edition, it now makes it possible for mainstream engineering companies to enter the world of High Performance Computing (HPC). This is clearly not a dig on Linux computing farms. I am a huge Linux fan, writing this post directly from my Linux netbook.

So why is Microsoft a game changer? Simple answer- they are familiar to people. But there is more to the story- the hardware needed is becoming mainstream. So it is really a combination of the two. Any engineer or IT professional can now setup a “cluster” or a “computing engine” and take advantage of the power in numbers.

So, here are two scenarios that you can consider when looking at hardware for your computing engine.

You can look at purchasing 2, 3 maybe 4 workstations or (equivalent servers) from Dell or HP or whoever. You will need to ensure that they have an infiniband connection to hook the two together. Fancy word for saying that each machine has a card and special cable for the processors to communicate to each other. Don’t bother trying to connect via Ethernet, you’ll use Ethernet to put them on your network so that you can communicate to them from a remote machine (laptop, workstation etc). Connecting 2 machines would require a cable, more than 2, you will need a “switch”. The more you want the more it will cost, BUT the quicker the solve, the bigger the return.

Here are a few pics of my multiple machine computing engine.

Here is the front- obviously.


Here is a look at the back. Pointing to the two infiniband connections (not ethernet) “hooking” the machines together to form the cluster. This whole configuration is in the $10k price range. This specific configuration are Dell Precisions with dual dual core processors and 8GB RAM each. The setup is fairly straightforward and is worth the effort for the cost.

But, if you have a bit deeper pockets, not super deep – but deeper. You could go with the “cluster in a box”. This is literally, unpack it and plug it in. Really nice. The “infiniband” type connections are all built into the box. In this configuration, each computer is referred to as a “blade”.


Here is a shot of the back of the box. Each one of the “slots” with a green light is the individual blade(computer). This box has 5 blades in all. The beauty is that this works as 2 clusters – one with 3 and one with 2 computers each. Essentially, the same speed/config as above.

Extremely exciting times right now for Design Engineers that are currently in or thinking about getting into the Upfront CAE space. There is massive potential for doing way more than we have ever been able to do before.

Personally, I am anxious to see where the MCAD world, MCAE space in general, finds itself leveraging various forms of cloud computing in the near future.