http://bugzilla.suse.com/show_bug.cgi?id=904517 Martin Pluskal changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |mpluskal@suse.com -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.suse.com/show_bug.cgi?id=904517 Robert Milasan changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |rmilasan@suse.com --- Comment #2 from Robert Milasan --- No, the attribute rotational is not always setup properly and will set deadline to possible other type of disks or not set it to all ssd. This is only an option for those who have ssd and wanna optimize their machines. The rule itself could be added to some optional package which deals with ssd disks (I have no idea if such a package exists). -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.suse.com/show_bug.cgi?id=904517 Stephan Kulow changed: What |Removed |Added ---------------------------------------------------------------------------- Assignee|bnc-team-screening@forge.pr |ihno@suse.com |ovo.novell.com | --- Comment #4 from Stephan Kulow --- I don't think something like that exists yet, but we have a PrjMgr for Storage... -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.suse.com/show_bug.cgi?id=904517 Duncan Mac-Vicar changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |mge@suse.com Flags| |needinfo?(rmilasan@suse.com | |) --- Comment #5 from Duncan Mac-Vicar --- Robert. What kind of disks have the rotational attribute not set correctly? Can you give an example? What is the probability of this miss-match, and how the consequences weight by setting the non-optimal scheduler for those cases compared to setting always the wrong one for all SSD users? Unless that package is installed automatically for users with SSDs having this not setup automatically is not very nice for a distro like openSUSE (I would understand if you had to do it manually for Gentoo or ArchLinux). If it is not possible using the rotational attribute, just matching for "Samsung SSD" should give us an improvement: Samsung has 25% of the market. -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.suse.com/show_bug.cgi?id=904517 --- Comment #7 from Dr. Werner Fink --- (In reply to Robert Milasan from comment #6) That is what I've seen after searching for 60-ssd-scheduler.rules. Maybe a simple script udev-generate-ssd-rule.sh similiar to udev-generate-persistent-rule.sh would be an option ? -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.suse.com/show_bug.cgi?id=904517 --- Comment #9 from Dr. Werner Fink --- (In reply to Robert Milasan from comment #8) Indeed this was my first thought ... somethink like /usr/lib/udev/udev-generate-ssd-rule /dev/sda checking for solid state disk /dev/sda found solid state disk /dev/sda create /etc/udev/rules.d/60-ssd-scheduler.rules adding rule for /dev/sda to /etc/udev/rules.d/60-ssd-scheduler.rules -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.suse.com/show_bug.cgi?id=904517 --- Comment #11 from Dr. Werner Fink --- (In reply to Robert Milasan from comment #10) Hmmm ... we may ask the caller of that script which may indicate this script as an interactive script? That is if the script may check for not rotational and if not true ask the caller what to do (skip, yes, no). -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.suse.com/show_bug.cgi?id=904517 Dr. Werner Fink changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |mc@suse.com Flags| |needinfo?(mc@suse.com) --- Comment #13 from Dr. Werner Fink --- (In reply to Michael Calmer from comment #12) Then explain why other distributions do not support /usr/lib/udev/rules.d/60-ssd-scheduler.rules but only explain how to add a local /etc/udev/rules.d/60-ssd-scheduler.rules? -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.suse.com/show_bug.cgi?id=904517 --- Comment #15 from Robert Milasan --- If rotational attribute would be 100% accurate, then the rule would 100% good and we would include it, otherwise is only a optimization thing which a user can do in case he/she knows all the details about it's own hardware. This 'tweak' could be included in the openSUSE wiki or how-to pages. -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.suse.com/show_bug.cgi?id=904517 --- Comment #17 from Robert Milasan --- Ihno, this is not a bug, this is a 'wish' which no distro is implementing. Why is this P3 ? -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.suse.com/show_bug.cgi?id=904517 Robert Milasan changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |mgalbraith@suse.com Flags| |needinfo?(mgalbraith@suse.c | |om) --- Comment #19 from Robert Milasan --- I'm not sure you can say 'wrong scheduler' and yes, now a days ssd are more and more popular, but still is an optimization thing to switch to a different scheduler. @Michael: what do you think? is switching the scheduler for an SSD that important or that good? I know it can have benefits, but I don't believe implementing a udev rule would be a good idea. Could be, maybe a tuning tip. What do you think? -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.suse.com/show_bug.cgi?id=904517 Jan Kara changed: What |Removed |Added ---------------------------------------------------------------------------- Flags|needinfo?(jack@suse.com) | --- Comment #22 from Jan Kara --- So here are my thoughts: 1) I agree that deadline is better for SSDs than CFQ (default). 2) I agree that 'rotational' attribute isn't 100% reliable but it's not completely random either. Usually when rotational == 0, you can be reasonably sure that standard optimizations for rotating disks don't apply. 3) If we set 'deadline' IO scheduler by mistake, nothing catastrophical happens. Just user may observe worse performance in some (not so usual) cases. So overall I believe using the udev rule is a net win (on average it should improve performance). There's one case I'm unsure about - when kernel parameter elevator=<foo> is used, user has probably thought about IO schedulers and decided that <foo> is the right default for him. In that case I guess we shouldn't interfere by changing the elevator in the udev rule... -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.suse.com/show_bug.cgi?id=904517 --- Comment #24 from Jan Kara --- The rule looks good to me. -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.suse.com/show_bug.cgi?id=904517 Arvin Schnell changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |aschnell@suse.com -- You are receiving this mail because: You are on the CC list for the bug.