4 Replies Latest reply on Aug 22, 2011 4:13 PM by Alec Chalmers

    Page Files

    Alec Chalmers

      I am running a fairly big frequency study with about 2 million nodes on a 64 bit Xeon system with 32GB of RAM and Windows XP 64 (which may be the problem) and SW 2011 SP 3.1. I let it run overnight (about 15 hours) and when I checked in the morning I had the message that virtual memory was being used and to expect "delays". Checking the performance tab in Task Manager I noticed that the peak commit charge had exceeded the physical memory at some point but was not presently exceeding the limit. Checking the harddisk I noticed the the entire 35GB had been dumped to the harddisk and almost completely filled the C drive, bringing the system to a crawl. As soon as I terminated the simulation, the 35GB disappeared from the harddrive and the systtem returned to normal speed (after a re-boot). I did a search for topics related tto this issue on this forum and found similar issues in earlier versions of SW, has anyone experienced this with a similar system? Does Windows 7 handle memory better?

        • Re: Page Files
          Paul Kellner

          How many cpus were involved in the simulation when it was running?

           

          Which solver were you using for modal analysis?

          How many mode shapes where you trying for?

          Why 2 million dof? Could you simplify?

          In Cosmos/M you could tell the software which hard drives to use for swap.

            • Re: Page Files
              Alec Chalmers

              I have a computer with dual Xeons (8 CPUs), but I don't believe SW Sim takes advantage of multiple processors like Flow Sim. I have simplified the model so that it will run without paging. When the application paged out I was running the Sparse solver because I tried using remote loads (which I have eliminated). I have been able to get results requesting modal information for 500 frequencies and still don't get the required .80 mass participation (getting about .70). When I try to run a random vibration linear dynamic analysis I keep getting a message asking for more points added to the base excitation curve, even though this same curve has run successfully on simpler models. Any ideas?