3 Replies Latest reply on Mar 16, 2010 3:38 PM by 1-L7XAXT

    File size handling

    Peter Holmi

      Does anyone have suggestions for running studies on large assemblies when the results take several minutes to appear?  I have run multiple simulations on the same assembly and the file is so slow to respond as to be almost unusable.  The studies include three shock tests (linear dynamic analysis (Modal Time History) studies, a static test, a frequency test, and three non uniform Random Noise vibration test (5-500 Hz, 10 Hz increments).  The model has 35K elements and 65.5K total nodes.  Each Random Noise test resulted in a 9.4 GB CWR file.  The simulation folder size is 79 GB.  It can take 20 minutes to switch between one result file and another.  During that time, the CPU usage remains below 10% the entire time and only runs on one of the cores.

      I am open to suggestions as to the best way to handle these studies.

        • Re: File size handling

          Hello Peter,

           

          Yes, I also run large studies and the results take a long time to appear.  I made a post about it titled "How Fast Should SW Simulation Display Results?"  Some people made interesting comments to me in that post.  I think the biggest reason that my studies are slow to display results is that my processor is fairly old now.  I have also watched my cpu usage and not noticed any horrendous disk activity.  However, the results display thing has to render all the colours onto all the faces in the model based on calculations done on the results data.  So I think it's a lot of start, stop, read, write calculations, and not one giant continuous process.  A cpu with a massive on-die cache would probably help a lot (but I can't say for sure, and they are expensive).  The new technology gets cheaper every year, so maybe it's worth checking out.

           

          Your models are titanic compared to mine, so it doesn't surprise me that your computer would take some time to display results.  Your .CWR file is 9Gb?  Mine are about 80Mb.  So even if you have a newer computer, 10 times faster than mine (5 years old now) there is still another order of magnitude difference in file size.  You say you are waiting 20 minutes between results files, well that is similar to what I wait if I want to see the Factor of Safety plot for my static study with 220,000 degrees of freedom (73k elements, 37k nodes).

           

          I sympathize with your frustration.  It would be really nice if the computer could work as fast as you could think, right?

           

          Since you have a newer computer, though, you can get done about 10 times as much work as I can in the same time.

           

          I suppose it's not much consolation to say "it could be worse!" but you can also check out my other post and see what those people had as ideas for me.

           

          Good luck, and lots of patience,

           

          David Fletcher

            • Re: File size handling
              Peter Holmi

              David,

              Thanks for your response.  Unless anyone knows better, I believe that your answer is correct.  I have since readjusted my model to get better mesh and frequency resolution versus what SolidWorks will handle.  The .CWR goes out to 45GB per test.  The good thing is that SolidWorks handles the simulation and the results appear to be valid.  It takes time; however, the end result is that I can get a viable result.  If I increase the number of frequencies tested or refine the mesh further, SolidWorks will no longer work properly and require me to stop the study.  For example, the elapsed time meter was incrementing hours like seconds.

              Thanks again for your response.

              Peter Holmi

                • Re: File size handling

                  One of the things I noticed when running simulations (in this case linear dynamic/modal time history) was that there are two main files that SW generates during the normal analysis runs:

                   

                  .STP files

                  .LCP files

                   

                  When the analysis finishes, the rotating green icon in the simulation tab keeps spinning, sometimes up to hours, before you can view the results.  I noticed that during this time SW is writing to the .CWR at the bristling pace of 4mb/sec.  Right now I'm waiting on an analysis that has an 18.4gb .STP and a 1.74GB .LCP.  I've also noticed that the results aren't viewable until the .CWR reaches a little more than the size of the .STP and .LCP added together.

                   

                  So, right now that total is 20.1GB.  My CWR file is currently 14.4GB.  This means it still has to write 5.7GB of data, and at 4mb/s that equates to waiting about 24 more minutes.

                   

                  During this time I see minimal ram usage and my 8 core CPU is wavering between 0-1% usage.  It doesn't seem that Simulation effectively utilizes the power available.  4MB/sec write speed is absurdly slow by today's standards unless I'm misunderstanding the task at hand.