CAS-W

[Anonymous]

Hello,

We are currently evaluating a hardware upgrade to support our Simulation CFD software. One of the options on the Dell site is a CAS-W optional cache solution to increase speeds. Have you heard of this? If so, have you heard of any problems using this with Simulation CFD?

Thanks!

[waited]

Daryn,

I have read about this technology, but I have not tested it using Simulation CFD.  You might be interested in this link, http://www.innova-systems.co.uk/assets/files/Dell%20Precision%20Workstation%20SolidWorks%20eGuide%20...

It shows performance gains using CAS-W.  For SolidWorks, they show only a 6% gain in IO performance.  Since this option mimics a solid state drive, the gains you get are directly proportional to the amount of IO you are doing.  Simulation CFD does not do a large amount of IO during solver execution.  However, the one exception to that would be an analysis where animation data was being save at a high frequency, which would generate many gigabytes of data.  I have run cases that generate over 100 gigabytes of the .res files.  For these cases, there should be improvements with a faster IO solution.  Maybe other Simulation CFD users with solid state drives can weigh in and say how much improvement they have seen with SSD while running Simulation CFD.  CAS-W claims to be competitive with SSD, so input from SSD users should be helpful in your evaluation.

Dave

[Anonymous]

Hi,

From my experience, SimCFD runs as a service in the background.  If the model gets so big that it exceeds memory capacity the program stops and warns (crashes).  As far as I know, it has no way of using virtual memory  (I am not a great programmer).   I went big on memory (64gigs) and CPU (duel E5-2687w V2) with my money.  I have both an SSD and a large mechanical hard drive.  Originally I ran solely on the SSD and quickly ran out of capacity.   Now the SSD runs the program and the mechanical receives the output.  Post processing is a bit slower.  When I run a large model the job could last several days.  Any time savings on data storage / retrieval would be in the noise level and hardly measurable from a return on investment prospective.  Having SSDs in Raid(x) would be nice but for big jobs going HPC with several computers in an array has been my thinking.  There must be an ideal ratio of CPU capacity to memory capacity but I do not know what that would be.

A program which moves vast quantities of data rather than the same data over and over would benefit more.  Think NSA!

Have a blessed day,

Mike

[Anonymous]

Thanks for the reply to my CAS-W question. 

"There must be an ideal ratio of CPU capacity to memory capacity but I do not know what that would be."  I have read that 4-8GB of physical RAM per core is recommended by COMSOL.  I imagine the same "rule of thumb" would work for simCFD.

What size SSD drive were you using when if filled up?

[Anonymous]

Thanks!

[Anonymous]

Hi Daryn,

My SSD is 238 gigs.and my system has 4 gigs per CPU.  This seams to be an adequate amount of memory for my computational capacity....

Note my 1.8 terabyte mechanical is over three quarters full.  Output is voluminous!

Mike