Do we have an ETA on the fixes for the VM problem that was said to be fixed in 2.4.14pre6.. I just watched 38M of swap get eaten for no apparent reason? Thanks. -----=====-----=====-----=====-----=====----- Ben Rosenberg mailto:ben@whack.org -----=====-----=====-----=====-----=====----- "Technological progress is like an axe in the hands of a pathological criminal" -AE
Why don't you have a go at tuning it to see if you can get better results. Have a read of /usr/src/linux-2.4.4.SuSE/Documentation/sysctl/vm.txt or whatever the fully qualified path is on your version. This will tell you what tunable options exist in /proc/sys/vm and what they are used for. The one thing I don't like about the swapon command on Linux is that it doesn't distinguish between swap used and swap reserved. what you may be seeing is reserved swap - which is basicly swap allocated for use 'just in case' the program that asked for it, needs it. Is your increase in swap usage actually showing up as swap out I/O operations when monitored by vmstat? If not, then I pressume that this is reserved swap, in which case it's not going to have any impact on system performance as real program data is not actually getting swapped out at all. According to the above document, "overcommit_memory" is the default behaviour, so setting it to 1 to turn it off (sometimes called lazy swap) should stop reserved swap allocation from being registered as used swap by the swapon command. On Friday 02 November 2001 2:22 am, Ben Rosenberg wrote:
Do we have an ETA on the fixes for the VM problem that was said to be fixed in 2.4.14pre6.. I just watched 38M of swap get eaten for no apparent reason?
Thanks.
-----=====-----=====-----=====-----=====----- Ben Rosenberg mailto:ben@whack.org -----=====-----=====-----=====-----=====----- "Technological progress is like an axe in the hands of a pathological criminal" -AE
Why the hell is everyone so concerned about the use of some swap space? Is the computer thrashing the disk? Has performance dropped to an unbearable level? Just because the system reports that some swap has been used or allocated, does not mean that the swap space is active - Have you looked at vmstat? Like many things on Linux, once a resource is allocated, it is not free'd... it just remains allocated. If the the system was thrashing the disk, then I would say there was some cause for concern. Quite frankly, with the VM that was used from 2.4.0 to 2.4.9, I was alsways a bit concerned about the apparent lack of swap use, as once the real memory was consumed the system would become unuseable as it thrashed to try and find resources. The OOM Killer is a great concept.. unfortunately, it doesn't work. I personally, much prefer the model of 2.2.x, unfortunately I need some of the features of 2.4, so I welcome the VM rework that has been done in 2.4.20, and from the testing I've done on heavily used NFS servers, it is much more stable... even if it does claim some swap has been used with free memory left. Of course this is just my opinion... -Herman Ben Rosenberg wrote:
Do we have an ETA on the fixes for the VM problem that was said to be fixed in 2.4.14pre6.. I just watched 38M of swap get eaten for no apparent reason?
Thanks.
-----=====-----=====-----=====-----=====----- Ben Rosenberg mailto:ben@whack.org -----=====-----=====-----=====-----=====----- "Technological progress is like an axe in the hands of a pathological criminal" -AE
-- To unsubscribe send e-mail to suse-linux-e-unsubscribe@suse.com For additional commands send e-mail to suse-linux-e-help@suse.com Also check the FAQ at http://www.suse.com/support/faq and the archives at http://lists.suse.com
God... I need to learn how to type... typos fixed below. Herman Knief wrote:
Why the hell is everyone so concerned about the use of some swap space? Is the computer thrashing the disk? Has performance dropped to an unbearable level? Just because the system reports that some swap has been used or allocated, does not mean that the swap space is active - Have you looked at vmstat? Like many things on Linux, once a resource is allocated, it is not free'd... it just remains allocated. If the system was thrashing the disk, then I would say there was some cause for concern. Quite frankly, with the VM that was used from 2.4.0 to 2.4.9, I was always a bit concerned about the apparent lack of swap use, as once the real memory was consumed the system would become unuseable as it thrashed to try and find resources. The OOM Killer is a great concept.. unfortunately, it doesn't work. I personally, much prefer the model of 2.2.x, unfortunately I need some of the features of 2.4, so I welcome the VM rework that has been done in 2.4.10, and from the testing I've done on heavily used NFS servers, it is much more stable... even if it does claim some swap has been used with free memory left. Of course this is just my opinion...
-Herman
Ben Rosenberg wrote:
Do we have an ETA on the fixes for the VM problem that was said to be fixed in 2.4.14pre6.. I just watched 38M of swap get eaten for no apparent reason?
Thanks.
-----=====-----=====-----=====-----=====----- Ben Rosenberg mailto:ben@whack.org -----=====-----=====-----=====-----=====----- "Technological progress is like an axe in the hands of a pathological criminal" -AE
-- To unsubscribe send e-mail to suse-linux-e-unsubscribe@suse.com For additional commands send e-mail to suse-linux-e-help@suse.com Also check the FAQ at http://www.suse.com/support/faq and the archives at http://lists.suse.com
-- To unsubscribe send e-mail to suse-linux-e-unsubscribe@suse.com For additional commands send e-mail to suse-linux-e-help@suse.com Also check the FAQ at http://www.suse.com/support/faq and the archives at http://lists.suse.com
* Herman Knief (herman@knief.net) [011101 19:35]: ->Why the hell is everyone so concerned about the use of some swap space? Because as stated before "This is a bug!" ->Is the computer thrashing the disk? Yes! ->Has performance dropped to an unbearable level? Yes, a large chunk of memory is suddenly dumped to swap and the system slows to a crawl. ->Just because the system reports that some swap has been used or allocated, ->does not mean that the swap space is active. Not "some" 30-140+ megs gets shoved to swap for NO REASON that us non-programmers can figure out. ->Have you looked at vmstat? No, I watch system usage via Gkrellm. ->Like many things on Linux, once a resource is allocated, it is not free'd... ->it just remains allocated. If the the system was thrashing the disk, then ->I would say there was some cause for concern. Well, this is true to a point, but if you have 100+ megs of FREE ram then it shouldn't take that and leave swap alone until it's needed. And once RAM is freed then swap can be flushed and it does flush..I've watched it do so. ->Quite frankly, with the VM that was used from 2.4.0 to 2.4.9, ->I was alsways a bit concerned about the apparent lack of swap use, as ->once the real memory was consumed the system would become unuseable as ->it thrashed to try and find resources. Well, you shouldn't be concerned. This is how Solaris and many other Unix OS's work. Only Wintendo will use memory in this fashion and it's a bad design. Unix uses ram first then swap because it's always faster to execute operations in RAM then on the harddrive. As I have said to many, many people over the years on this list.. You should take Yoda's advice and "Unlearn what you have learned" because it's just plain wrong and I hope Unix/Linux never, ever starts to do this. But we know from the release notes that the problem surfaced in .10 and was fixed in .14pre6. There were several releases over the past month so it's not like the issue existed for years and is now being changed. It was recognized as a bug and has been fixed. We are just waiting for SuSE to test a kernel with the fix and release it which should thankfully be soon I hope to God... And I have never had a Linux/Unix system thrash because it ran out of memory and couldn't find it's swap if it needed it. Unix and Linux are much better at recovering memory from programs that have been closed then any Microsoft OS has ever been. Also, a side note...I'm not going off on MS just to do it...and don't explain how Microsoft OS's work. Before I started working with Unix I use to be a quite skilled NT Admin..so I know what I'm talking about. ->The OOM Killer is a great concept.. unfortunately, it doesn't work. ->I personally, much prefer the model of 2.2.x, unfortunately I need ->some of the features of 2.4 Well, can you point out when the 2.2.X kernel opted for swap over physical memory? The first kernel I used was 1.1.13 and just about every 2.2 kernel up to 2.2.18 when I switched to the 2.4 kernel with the .2 release. I've never see this behavior... -> so I welcome the VM rework that has been done in 2.4.20, and from the ->testing I've done on heavily used NFS servers, it is much more stable... ->even if it does claim some swap has been used with free memory left. 2.4.20 doesn't exist :) ..they are working on 2.4.14 hence the pre6 just being released. I'm a systems admin for 4 large Solaris clusters that use NFS quite a bit to run some 600,000 domains and they do not behave like you discribe..if they did I would look for a patch on Sun Solve. :) ->Of course this is just my opinion... True..and hopefully we can educate you on the Unix/Linux way :) Cheers! And remember have fun! -----=====-----=====-----=====-----=====----- Ben Rosenberg mailto:ben@whack.org -----=====-----=====-----=====-----=====----- "Technological progress is like an axe in the hands of a pathological criminal" -AE
On Thu, Nov 01, 2001 at 10:17:21PM -0800, Ben Rosenberg wrote:
But we know from the release notes that the problem surfaced in .10 and was fixed in .14pre6. There were several releases over the past month so it's not like the issue existed for years and is now being changed. It was recognized as a bug and has been fixed. We are just waiting for SuSE to test a kernel with the fix and release it which should thankfully be soon I hope to God...
Ben, please don't take me wrong. This is the changelog from 14pre6. Which change fixes your problem? pre6: - me: remember to bump the version number ;) - Hugh Dickins: export "free_lru_page()" for modules - Jeff Garzik: don't change nopage arguments, just make the last a dummy one - David Miller: sparc and net updates (netfilter, VLAN etc) - Nikita Danilov: reiserfs cleanups - Jan Kara: quota initialization race - Tigran Aivazian: make the x86 microcode update driver happy about hyperthreaded P4's - me: shrink dcache/icache more aggressively - me: fix up oom-killer so that it actually works
2.4.20 doesn't exist :) ..they are working on 2.4.14 hence the pre6 just being released.
That was an obvious typo. Sometimes fingers become too fat ;-) And actually 14pre7 is out: Changelog: pre7: - me: reinstate "delete swap cache on low swap" code - David Miller: ksoftirqd startup race fix - Hugh Dickins: make tmpfs free swap cache entries proactively You also mentioned kernel 1.1.13. Should we read it 1.2.13? Best regards, -Kastus
Ok..ok...some typos were made..but I won't take back the bug exists. :P -----=====-----=====-----=====-----=====----- Ben Rosenberg mailto:ben@whack.org -----=====-----=====-----=====-----=====----- "Technological progress is like an axe in the hands of a pathological criminal" -AE
* Herman Knief
concept.. unfortunately, it doesn't work. I personally, much prefer the model of 2.2.x, unfortunately I need some of the features of 2.4, so
The model Linus incorporated from Andrea Arcangeli resembles 2.2.x quite a lot, so run a kernel >= 2.4.12 -- Mads Martin Joergensen, http://mmj.dk "Why make things difficult, when it is possible to make them cryptic and totally illogic, with just a little bit more effort." -- A. P. J.
On Friday 02 November 2001 01:34 pm, Mads Martin Joergensen, went on about:
* Herman Knief
[Nov 02. 2001 04:34]: concept.. unfortunately, it doesn't work. I personally, much prefer the model of 2.2.x, unfortunately I need some of the features of 2.4, so
The model Linus incorporated from Andrea Arcangeli resembles 2.2.x quite a lot, so run a kernel >= 2.4.12
So Mantel and SuSE both use the Linus version of the kernel and not the Alan Cox version? I understand the Linus version uses the newly rewritten VM which fixes a lot of things, but I don't remember at which kernel revision he started using it, seems like 2.4.10 and up is the new fixed version of VM, is that right Mads? Regards O'Malley -- ---KMail 1.3.1--- SuSE Linux v7.2--- Registered Linux User #225206 /tracerb@sprintmail.com/ *Magic Page Products* *Team Amiga* http://home.sprintmail.com/~tracerb
* Lee O'Malley
On Friday 02 November 2001 01:34 pm, Mads Martin Joergensen, went on about:
* Herman Knief
[Nov 02. 2001 04:34]: concept.. unfortunately, it doesn't work. I personally, much prefer the model of 2.2.x, unfortunately I need some of the features of 2.4, so
The model Linus incorporated from Andrea Arcangeli resembles 2.2.x quite a lot, so run a kernel >= 2.4.12
So Mantel and SuSE both use the Linus version of the kernel and not the Alan Cox version? I understand the Linus version uses the newly rewritten VM which fixes a lot of things, but I don't remember at which kernel revision he started using it, seems like 2.4.10 and up is the new fixed version of VM, is that right Mads?
Yes, that is correct. The SuSE kernel is of course also enhanced with extra stuff from SuSE Labs. -- Mads Martin Joergensen, http://mmj.dk "Why make things difficult, when it is possible to make them cryptic and totally illogic, with just a little bit more effort." -- A. P. J.
participants (6)
-
Ben Rosenberg
-
Herman Knief
-
John McNulty
-
Konstantin (Kastus) Shchuka
-
Lee O'Malley
-
Mads Martin Joergensen