Feature changed by: Jan Kara (jankara) Feature #305168, revision 23 Title: default to relatime openSUSE-11.1: Rejected by Stanislav Visnovsky (visnov) reject date: 2008-09-10 10:20:17 reject reason: This is too late for openSUSE 11.1. Postponing. Priority Requester: Mandatory openSUSE-11.2: Evaluation Priority Requester: Mandatory Requested by: Michael Meeks (michael_meeks) Description: At least for desktops, we should reduce the memory and I/O thrash caused by atime updates, by using relatime. Relations: - bugreport from community (novell/bugzilla/id: 398616) https://bugzilla.novell.com/show_bug.cgi?id=398616 - Offers 'noatime' when 'relatime' is clearly better (novell/bugzilla/id: 461829) https://bugzilla.novell.com/show_bug.cgi?id=461829 Discussion: #1: Robert Davies (robopensuse) (2009-01-16 15:52:09) Bug #461829 is not requesting breaking POSIX semantics by default! relatime should replace noatime option in the mount options part of partioner tool, especially now it has been improved to update atime on file access if it's older than 24 hours. Standards are important, and LRU file caches cannot be relied on if noatime usage is common, or if poor backup software touches atimes, when reading files. #2: Sven Burmeister (rabauke) (2009-01-23 01:09:47) 11.1 is out and 11.2 getting started. Did anyone hear of ubuntu suffering from using relatime by default? If not, it seems likely that there are little to none issues. #3: Stephan Kulow (coolo) (2009-05-08 16:51:21) Matthias, I would think the easiest solution is replacing the current checkbox "use noatime" with "use relatime" and make it default (possibly per product). atime is a major pain for many systems where powersave and co are affected. #4: Matthias Eckermann (mge1512) (2009-05-11 15:52:11) (reply to #3) I am not perfectly happy, as a system crash could invalide the "atime"s with "relatime" switched on; looking for supporting information though, it seems, the broader Linux community is in favour of "relatime" (with remarkable exceptions); thus I propose: * introduce "relatime" as an independend parameter * keep the "noatime" checkbox in YaST * make "relatime" default on SSDs automatically * introduce a generic switch in YaST "defaultrelatime", which can be switched on/off per product * make "relatime" default "on" for the openSUSE product Does that help? #5: Jan Kara (jankara) (2009-05-11 16:33:11) This is a bit separate to how it will all look in Yast but I wanted to point out that in current Linus's tree (going to be 2.6.30) is a patch which changes kernel's default to 'relatime'. So unless we change something, all filesystems on this or newer kernel are going to be mounted as relatime (you have to specify 'strictatime' mount option if you really wish to keep atime). #6: Stephan Kulow (coolo) (2009-05-12 10:36:27) (reply to #5) then we can leave everything as it is actually and the feature is Done once we switch to 2.6.30 #7: Jan Kara (jankara) (2009-05-12 13:08:05) (reply to #6) Yes, the original request will be fulfilled. But I believe it would be good to have a selection in partitioner tool like: "Access time behavior: <relatime|noatime|strictatime>" anyway and make 'relatime' default. Also it is a question how to behave on update - the safest would be to add 'strictatime' to filesystems that don't have any atime setting but realistically it is probably better to ask user whether he agrees to switching to relatime and default to "yes". #9: Stephan Kulow (coolo) (2009-05-12 13:45:51) (reply to #7) completely different feature IMO and way more complex (and definitely over the top of openSUSE 11.2) #10: Duncan Mac-Vicar (dmacvicar) (2009-05-12 14:00:25) (reply to #7) Do we really need this extra options really? Is there a good usecase for it? I don't want to bring something to implementation because it is "good", "cool" or "nice to have". + #11: Jan Kara (jankara) (2009-05-12 15:50:33) (reply to #10) + User can always edit /etc/fstab by hand so strictly speaking it's not + necessary. About the usecase - there really are people who want + 'strictatime' for various reasons (well, most often they are just + afraid that something can break). From my POV that is enough for + substituting that 'noatime' checkbox we have currently with the + selection I suggested but I admit I'm not the one who'll have to + implement it ;-). + Regarding the update - silently changing the behavior of atime seems + kind of rough to users to me. At least we should warn them (release + notes?) that this is happening and that they can add 'strictatime' to + /etc/fstab if they have problems with the change. -- openSUSE Feature: https://features.opensuse.org/305168