Saturday, February 07, 2009

Trends of Computing Power in the Upfront CAE Industry

Since the beginning of time, Upfront CAE engineers have always been pushing the limits of computing power. The CAE industry has always been leading the charge in maximizing computer resources in attempt to solve larger and more sophisticated problems. Even in the early days when solving a simple 2D stress analysis, the computing power required was much greater than what was available to the average user.


The battle is on two fronts: "get more with less" and "get more with more".

GET MORE WITH LESS

There are some of us in the world that are using workstation class laptops, such as the Dell M6300. This is a powerhouse machine that is capable of handling fairly large CAD models and can run fairly large CFD models. The benefits of a workstation laptop are pretty obvious: ability to have all engineering AND office applications all on one machine, with the ability to take your work on the road or simply having the ability to work at home/remote etc.


So what are the negatives???
  • Machines and power supplies are a bit bulky and HEAVY- not the most ideal for traveling
  • Battery life is SUB-par even for basic office applications; near useless for CAD or CFD work
  • Durability is questionable- I have had about 7 of these machines and everyone has had hardware failures in less than 2 years
  • The "all-in-one" machine becomes unusable when a simulation is running.
  • There is a risk to having your number cruncher mobile. If the machine drops or is damaged, you risk losing valuable information that sometimes takes hours to days to compute.
Progressive companies that understand the need for speed, budget accordingly and provide engineers access to a second machine. Sometimes this secondary machine is shared. Some companies have grand-visions of having a computing "server". I like to call this the "machine in the corner". The idea is that users can work locally on their laptop or mortal desktop and have the big ole' "machine in the corner" to crunch the numbers. Many organizations pull this off, but many have come up short. Why is this?

The fact is until recently, the "machine in the corner" was expensive and could do more than the laptops or smaller desktops but were always plagued with out-dated memory, limited access when outside the company firewall and unreliable means of interacting with the "machine in the corner" efficiently. So, what inevitably happened was that machine was often monopolized by one or two guys and the machine would just crank away for hours. Or WORSE, the investment is made and the only thing happening in the corner is dust collecting.

GET MORE WITH MORE

This methodology is what I refer to as maximizing the investment. Just when we were seeing a slow down in the "machine in the corner" way of life, it is having a bit of a resurgence.

Let's take a look at why...
  • People are "connected" more now than ever before
  • It is a common way of life in companies to provide access from "outside"
  • Computing power is increasing everyday
  • CAE vendors are automating sophisticated analyses; making things easy today that were not even possible only a few years ago
  • FEA and CFD are becoming mainstream; therefore the demand is increasing everyday
  • Engineers want to solve larger and more sophisticated models
  • CAE is becoming part of the design process, so the need to run multiple cases as part of a design study is becoming mainstream
  • Hardware prices are now within everyone's reach
  • Hardware performance is increasing everyday
But as the "machine in the corner" is becoming popular, so are the "MACHINES in the corner" OR the "cluster in the corner". But, don't panic, it is not your father's compute farm that I am talking about. Parallel computing, distributed computing or High Performance Computing has been around for quite some time and is heavily used by the elite companies in the world that have 16, 32, infinite number of Linux computers at their disposal to run models for their superstar analysts. These clusters are very sophisticated, require a dedicated team of analysts and IT professionals to maintain and utilize its immense power.


What I am talking about are simple, 2, 4, maybe 6 basic nice windows machines that are networked together using Windows Server 2008 and an Infiniband network connection that can run multiple models in a fraction of the time it would take on one machine. For organizations that are seeing the value of leveraging FEA and CFD simulation as part of the design process and have a need to do multiple "what-if" scenarios or have always felt limited in what could be run due to hardware limitations, can easily justify the cost of a few machines to speed up the process and deliver results that used to take all night. These models can now be run in a few hours.

So, let's say it takes on average 6 hours per simulation and even on your most efficient week you were able to run 2 models/night. You were getting ~10 simulations done a week or 10 what-if scenarios a week. Imagine if you had 2 dedicated machines making up your "cluster in the corner" and you were able to run 45 simulations a week. Or think of it a different way, what took you a week, you can now get done in a little bit more than a day. That payback on efficiency can easily justify the cost of the additional hardware and the software licensing required to access the increase in speed.

This idea of a plug and play cluster will be a game changer for those companies that have felt as if their hands were tied. This is the next generation way of thinking about Upfront CAE. Keep your eyes peeled for the trends from the various vendors.

It is no secret that CFdesign is very focused on High Performance Computing (HPC) and providing a means to solve larger models and more what if scenarios FAST. This doesn't mean that the doing "more with less" crowd will be left in the dark. The needs of the Upfront market are vast, and the technology out there is constantly changing. All of the chip maunfacturers are promising performance boosts in the near future. Quite frankly, the R&D that happens in the area of performance benefits everyone. Getting just a bit more from what we already have is just as important as anything.

1 comment:

Anonymous said...

I really appreciate that you talk about the ideal benefit of HPC for ufpront CAE, but also list the potential realities of what can happen with a poor implementation. Bravo.

Absolutely, if a company already has their upfront CAE ducks in a row, and is already getting a big return, they should go invest the tiny amount it take to setup a solver cluster. Because, it's highly likely their engineers will actually use it, and the ROI will be instant.

One thing Blue Ridge has always been good at is making software for the everyman. That gets real important in this HPC scenario. If it takes any more knowhow, tweaking, or button pushing for the part-time CAE crowd to "flip the switch on for an HPC run" than that cluster in the corner will gather dust. I'm betting Blue Ridge will deploy this without requiring much extra thinking on the customer side :)