Darac Marjal wrote: > However, as you've noted, once you run out of RAM, the kernel should > start moving the less-frequently used pages into Swap. In theory, the > OOM-killer should only come into play when both are full.
Agreed. > However, I can see a couple of situations where that may not happen: Note that the Bret said he was running on a 64-bit machine and that he had configured 40G of swap space. Therefore he won't be experiencing 32-bit memory limitations and he won't be experiencing out of virtual memory space. Instead a large process will simply grow larger. If the process walks through its own memory with any regularity then it will all be somewhat active which will cause it all to page in and out of swap space. That will definitely cause the machine to behave very slowly while it is actively swapping memory pages. > Bret Busby wrote: > > It used to work, much better, with Debian 3 and 3.1; I can't > > remember much about Debian 4, then, as previously mentioned, I had > > the problem and the solution as such, with Debian 5, and, now, with > > Debian 6, memory management appears to simply not work, making > > Debian 6, at least in the 64 bit version, of the nature of the > > attributes used to describe the experimental version of Debian. But remember that amd64 is a recent addition to Debian. Older Debian 3 and 3.1 did not have an amd64 64-bit version. (There were 64-bit alpha versions and so on.) Therefore it is likely that you were running a 32-bit version of the system back at that time. That is a critical difference. It is critical because then you would have been capped out at 2G per process (or 3G if they linked it -N non-shared to get that extra gig) and by that limitation. If you were running a 32-bit version of Opera it would have run out of address space for any more ram and could not possibly have grown to be a 14G process image as you found it to be recently. It is only possible now because you have a 64-bit system now. As far as the swap space goes... I imagine that you probably increased it due to the virtual memory pressure from Opera's large process size. Since it has been growing to that large size it is using up virtual memory. I assume that is why you increased to that large size of swap. Which is fine. Because otherwise being out of virtual memory the linux kernel would have activated the out-of-memory killer and it would have killed processes on your machine until it could pay its memory overcommit debt. (Depending upon the setting of vm.overcommit_memory.) I don't know why Opera is so large on your system. I think addressing it would be best place to improve your situation. Bob
signature.asc
Description: Digital signature