Hi Jord,
To my knowledge, Windows does not have a way to accurately determine how much memory is truly being used by a VM. And thus, it is a special case. VirtualBox states: https://www.virtualbox.org/manual/ch03.html#settings-motherboard 3.4. System settings 3.4.1. "Motherboard" tab Base memory This sets the amount of RAM that is allocated and given to the VM when it is running. The specified amount of memory will be requested from the host operating system, so it must be available or made available as free memory on the host when attempting to start the VM and will not be available to the host while the VM is running. David's solution to this, was to make BOINC budget that base memory amount, whenever the BOINC task scheduler schedules a VirtualBox VM task. Does the task VM's value for "base memory", match the BOINC task's property for "Working set size". It should, per David's code, I believe. Also, is that a problem somehow? Finally, how did the LHC tasks circumvent that - did they end up using more memory than the "base memory" somehow (dynamically grow?) and then BOINC didn't account for it? Regards, Jacob ________________________________ From: boinc_dev <boinc_dev-boun...@ssl.berkeley.edu> on behalf of Jord van der Elst <els...@gmail.com> Sent: Saturday, March 4, 2017 5:24 AM To: David Anderson Cc: BOINC Dev Mailing List Subject: [boinc_dev] Project application using more memory than it reports to BOINC Hi David, A user on the BOINC forums reported that he was now getting ready to get rid of BOINC, because it would use his whole memory and then even more, swapping to page file and there was nothing he could do about it. (ref https://boinc.berkeley.edu/dev/forum_thread.php?id=11494) He states to be running into this problem when running ATLAS GARL tasks, that use 2GB per task, which he appeared to run 10 of at the same time, using then 20GB of memory. A bit troublesome on a system using 16GB. So stopping that, he switched to LHC and running the LHCb simulation he now runs into to same problem. Here however there is a thread explaining why that is (https://lhcathome.cern.ch/sixtrack/forum_thread.php?id=4072): the LHCb tasks use 2GB of memory but report they only use 480MB per. So BOINC thinks they use 480MB, but they really do use 2GB. These all run in VirtualBox for as far as I understand, so perhaps that there's a bug in the wrapper how the memory usage is being read. In any case, shouldn't BOINC always poll the actual memory in use and stop using more when all of the available RAM has been consumed, even if there are free CPUs? Thanks, -- Jord van der Elst. _______________________________________________ boinc_dev mailing list boinc_dev@ssl.berkeley.edu http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev To unsubscribe, visit the above URL and (near bottom of page) enter your email address. _______________________________________________ boinc_dev mailing list boinc_dev@ssl.berkeley.edu http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev To unsubscribe, visit the above URL and (near bottom of page) enter your email address.
