Carles Pina i Estany <[EMAIL PROTECTED]> wrote: >> Everything about this topic is a opinion, but not some rules in some "old" Kernels (2.0 or 2.2, I remember that i have read some problems with less swap than RAM memory for some algorithms). But not today! <<
I think that pretty much sums it up - you can produce good rules only by taking into account: * The kernel version (does it have the OOM Killer patch) * The application working set in-memory and total requirements * The hardware configuration (large or small hard drive, or no hard drive at all, e.g. embedded system) >> mmm... it is not 100% correct if we have OOM Killer activated... (2.6 Kernels and I think that the last 2.4 Kernels, some 2.4 Kernels it was activated by default -you need Openmosix patch to disable this opinion) << Yes - I wasn't taking the OOM Killer patch into consideration - in fact, I originally wrote that some time before that patch was introduced. >> I think that sometimes Linux kill some process that is not the biggest... I am not 100% about which algorithms follows for this. << Again, depends on the kernel - from memory some versions kill the process making the request that's causing the OOM condition, others kill low-priority processes. The algorithm used has been a point of contention. >> good point to have more swap space to feel that something is not going fine, I like the idea. << I think it's a reasonable scheme for single-user (e.g. desktop workstation) systems. The best approach is to stay well away from unexpected (to the user) effects, whether it be processes being suddenly killed or the system freezing. A sluggish system with the drive activity light constantly flashing usually is enough to convince the user to back off before any "damage" is done. Of course, this can't work for multi-user systems, e.g. servers. >> Althouth, sometimes Linux uses swap space only to safe RAM from some sleeping (for long time) processes. So maybe we have a big process into the swap (so, free will say that we are using swap) and the system is really fast: everything that we are using is in RAM. << That's true, but I think that will be an unusual situation for the most common workstation and server environments, since those processes are things that usually load and initially run during system startup, so they usually fit quite easily in RAM and don't represent much space in the swap partition. If the swap partition is 3x RAM, and is being used heavily, it's almost certainly by runnable processes. I should probably add </opinion> here. ;) >> Thank you for your answer, and excuse me some differing opinions... maybe I don't know something or something that I know it is not in good way. << Thank *you* for raising it - you've made me revisit the topic and I'm going to add some more detail to my course materials. I think the picture that's emerging is that the topic is *not* a simple one, and that simple questions in the tests may not be appropriate. Hopefully others will have more to add - and we haven't even considered things like suspend-to-swap yet. . . Best, --- Les Bell, RHCE, CISSP [http://www.lesbell.com.au] _______________________________________________ lpi-examdev mailing list lpi-examdev@lpi.org http://list.lpi.org/mailman/listinfo/lpi-examdev
