Bug#620993: general: Lenny 2.6.26-2 has noticably increased swap usage, tho not swap thrashing
Package: general Severity: normal Swap usage prior to 2.6.26-2 upgrade was a relatively constant 0 Post 2.6.26-2 upgrade the usage is usually up into 300MB, not actively swapping however and doesn't seem to be affecting performance, but nagios is none to happy about swap being used. The problem seems relegated to a handful of machines, although there may be some usage pattern I'm not seeing but otherwise there is no difference between systems that are "swap happy" and those where swap stays 0, same make/model, same build configuration, packages, client code, etc. Like I mentioned, it doesn't seem to impact performance, but it is certainly a change from the norm. See this on about 8 systems out of a couple hundred, and the same 8 will almost always start using swap slowly over time until a plateu is reached of around 300-500MB, depending on the system. -- System Information: Debian Release: 5.0.8 APT prefers oldstable APT policy: (500, 'oldstable') Architecture: amd64 (x86_64) Kernel: Linux 2.6.26-2-amd64 (SMP w/8 CPU cores) Locale: LANG=C, LC_CTYPE=C (charmap=ANSI_X3.4-1968) Shell: /bin/sh linked to /bin/bash -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/20110405180845.13396.76364.report...@www05.wernerpublishing.ml.zerolag.com
Bug#620993: closed by Ben Hutchings (Re: Bug#620993: general: Lenny 2.6.26-2 has noticably increased swap usage, tho not swap thrashing)
I have, but fixing monitoring to suit edge cases created from a recent upgrade doesn't make the edge cases non-issues. This is still an issue whether you want to hide it under nagios or not, so I'd appreciate a little more assistance in finding the problem beyond "fixing nagios". If you go the doctors for a broken leg he doesn't just tell you "well try not to walk on it". The upgrades to 2.6.26-2 created anomalies, anomalies are bad. On 4/5/2011 11:48 AM, Debian Bug Tracking System wrote: This is an automatic notification regarding your Bug report which was filed against the general package: #620993: general: Lenny 2.6.26-2 has noticably increased swap usage, tho not swap thrashing It has been closed by Ben Hutchings. Their explanation is attached below along with your original report. If this explanation is unsatisfactory and you have not received a better one in a separate message then please contact Ben Hutchings by replying to this email. -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/4d9b6a40.1050...@gmail.com
Bug#620993: closed by Ben Hutchings (Re: Bug#620993: general: Lenny 2.6.26-2 has noticably increased swap usage, tho not swap thrashing)
I'm not arguing that, I fully expect the kernel to use it to swap *if needed*. And if this was 20MB of swap, or maybe 100MB, ok, sure, the kernel might be swapping old pages out, but 300MB+ swapping out in 2.6.26-2 where 0MB swapped out in 2.6.26-1, and I can't find anything in the kernel.org or debian kernel changelogs referencing a change along these lines doesn't really scream "working as expected". More annoying that the swap space never decreases unless swapoff/swapon occurs The closest thing I can find was a patch from Mel Gorman, but that doesn't explain how the swap space is getting used in the first place, or more how 2 otherwise identical machines have different amounts used, 0 and in the case I'm debugging ATM 312MB, Same make/model, same debian lenny build, they're clones for all intents and purposes. The only thing on these systems that has changed was 2.6.26-1 to 2.6.26-2, and swap wasn't being used to this extent, or staying used if it was used, prior to 2.6.26-2. Now it almost seems that once its used it doesn't get freed. On 4/5/2011 1:45 PM, Vincent Danjean wrote: On 05/04/2011 21:15, Daniel Gary wrote: I have, but fixing monitoring to suit edge cases created from a recent upgrade doesn't make the edge cases non-issues. This is still an issue whether you want to hide it under nagios or not, I do not understand the issue. You have some swap and you expect the kernel *not* to use it ? *This* would be a bug. It is better that the kernel swap out (parts of) processes it never uses and keep the RAM for processes that need it. Regards, Vincent -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/4d9b86fd.9000...@gmail.com
Bug#620993: closed by Ben Hutchings (Re: Bug#620993: general: Lenny 2.6.26-2 has noticably increased swap usage, tho not swap thrashing)
On 4/5/2011 2:47 PM, Russell Coker wrote: On Wed, 6 Apr 2011, Daniel Gary wrote: I'm not arguing that, I fully expect the kernel to use it to swap *if needed*. And if this was 20MB of swap, or maybe 100MB, ok, sure, the kernel might be swapping old pages out, but 300MB+ swapping out in 2.6.26-2 where 0MB Using more memory for disk cache and less for storing rarely used pages sounds like a performance optimisation. Fully agree, but I don't see anything in the changelog where that optimization occurred. So while it sounds great in theory I can't back it with the patches that went into 2.6.26-2 from 2.6.26-1. Although it is what I've been telling the client is the likely case for lack of a better answer and no loss in performance on the system. Anyway I don't think that there's going to be much interest in performance- tuning of Lenny kernels now that Squeeze is released. Unfortunately an upgrade to Squeeze is a bit more involved than the upgrade windows we have available, but I think may be the only option at this point, but that's kind of like fixing a sparkplug that doesn't fire by buying a new car, sure it'll probably do the job in the end, but that doesn't really explain why the sparkplug doesn't fire. But certainly a better reason to close the ticket than the previous one, doesn't make it invalid, just makes it a wontfix. I'd rather it be closed for someone being lazy than someone being dismissive, a good sysadmin is a lazy sysadmin after all. If you want help in tuning Lenny then probably debian-user or your local LUG mailing list would be the best option. Its not a tuning issue, its a change in a vacuum, more a freakish annoyance than anything, it performs fine, we benchmarked before & after and didn't notice any change either way other than these systems that swap, well... swap. The support tech in me says "screw it, it works", but the engineer in me says "hmm, thats funny". -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/4d9b9510.5060...@gmail.com
