Thermal analysis not using full potential of a single core?

[niles273]

I gather that most of cosmos/simulation is single threaded most of the time for a single simulation. Although I've come to terms with that, I'd still like to get everything I can out of a single thread. I've noticed that while I run simulations total CPU usage hovers around 6% and rarely exceeds 10% (except briefly between time steps when it goes truely parallel) There must be some sort of bottleneck somewhere causing this but I don't know where to look. Has anyone successfully tackled a similar issue? Is this the case for everyone?

A few details I left out in the below picture are that i was using the sparse solver (although both are the same for me in this regard) and that it was a transient analysis.

Thanks!

 

[kellnerp]

In the processes tab of task manager turn on the three disk I/O byte columns and the memory and page fault columns. See if there is a lot of disk activity or page faulting.

If there is a lot of disk activity you might benefit from a solid state hard drive. I haven't tested them yet, but for FEA with a lot of disk access they should help.

Of course the first thing to ask is how much ram you have. I'm guessing you have quite a bit, but I have to ask.

TOP
CSWP, BSSE

http://www.engtran.com

http://www.niswug.org

"Node news is good news."

 

[SteveMartin]

I'll throw out another 'should be obvious' check I've seen people miss before: If you have COSMOS set to store results on a network location it can spend a LOT more time sending info back and forth across the network than working the local machine - it doesn't just put the results there when the analysis is done. I don't know if this is improved/different in 2010 (I've seen this a recent as 2008, but don't have experience with more current versions).

I'm surprised it's only loading one core for your analysis. I haven't done much thermal (other than in FloWorks), but my static FEA runs can often utilize all eight cores on my machine.

 

[niles273]

DOH!!! Thanks Steve!

It took me a while to free up some time to check but when I saw the word server in the destination file I literally laughed out loud. I set it to a local file and it ripped through the above simulation in 42 seconds where it took 10+ minutes before. Still only running a single core for the bulk of it but going from 6% to 25% of available resources is a -huge- productivity boost.

I didn't immediately try it out because I was "so sure" we didn't have it set up that way. Now I can increase the complexity of my simulations an order of magnitude and get back to complaining

 

[kellnerp]

You also need to look at your network.

TOP
CSWP, BSSE

http://www.engtran.com

http://www.niswug.org

"Node news is good news."