[Date Index] [Thread Index] [Author Index]
MemoryInUse reports wrong numbers?
We are running MMA 2.1 on an HP9000/750 (HP-UX 8.07). When a MMA session gets fairly large ( >~ 8 Megabytes or so), MemoryInUse gives results which are greatly different from those reported by either HP's Glance, top, or vmstat(1), all three of which agree that MemoryInUse is returning a number about a factor of 2 too small. Note that I am comparing the resident program size from Glance/top/vmstat to MemoryInUse, not the total program size. Further, Glance/top/vmstat seem to be correct since the UNIX kernel begins paging when MMA has eaten all the RAM. As an undergraduate computing facility, we placed a governor on the size of a MMA session based on David Withoff's <withoff at wri.com> posted automatic invocation of Share for large kernels, supplemented by Todd Gayley's <tgayley at wri.com> extension invoking ClearInOut and CleanSlate if Share doesn't sufficiently reduce the size of the kernel, and further extended by me to Exit if those don't do the job. We now limit to MemoryInUse of 8000000 (8 Megs) to get an effective limit of 16 Megs. Is this a generic bug or HP specific? Am I misinterpreting something and it isn't a bug at all? Finally, could it be related to Jesus Rivero's recent post to this group of problems with big packages under Windows? -- David Harrison | "We place no reliance on virgin Dept. of Physics, Univ. of Toronto | or pigeon, Inet: harrison at faraday.physics.utoronto.ca | Our method is science, or aim Tel: 416-978-2977 Fax: 416-978-5848 | is religion." Aleister Crowley