| shadeslayer | hey ho | 10:12 |
|---|---|---|
| apw | cking, you did some testing for us, on cfq/deadline etc, as we were considering flipping back to cfq "because clearly it is better" ... and it wasn't | 10:13 |
| apw | cking, shadeslayer is asking to switch to cfq ... | 10:13 |
| cking | apw, as per usual, it depends on the usecase | 10:13 |
| shadeslayer | fwiw userland stuff depends on CFQ https://bugs.launchpad.net/bugs/1310402 | 10:14 |
| ubot2` | Ubuntu bug 1310402 in linux "Userland depends on ionice idle but default scheduler is "deadline". " [Wishlist,Triaged] | 10:14 |
| ubot5 | Ubuntu bug 1310402 in linux (Ubuntu) "Userland depends on ionice idle but default scheduler is "deadline". " [Wishlist,Triaged] | 10:14 |
| apw | shadeslayer, well you can simply change it via sysfs | 10:14 |
| shadeslayer | apw: I'm not arguing about that | 10:14 |
| shadeslayer | it's just that it seems like a runtime dep | 10:14 |
| apw | but as i recall the analysis we switched to deadline, because we saw visible delayes in desktop from cfq | 10:15 |
| shadeslayer | I see, atleast for Kubuntu we wanted to switch to CFQ because the file indexer was trashing the HDD since io niceness was not available on deadline | 10:15 |
| shadeslayer | so I kind of did this https://launchpad.net/ubuntu/+source/kubuntu-settings/1:14.10ubuntu3 | 10:16 |
| apw | which seems reasonable | 10:16 |
| cking | the main problem is that whatever one choses there will be cases where one ioscheduler is better than the other | 10:16 |
| shadeslayer | I'm not arguing against that :D | 10:17 |
| apw | i am not actually sure what you are arguing | 10:17 |
| shadeslayer | All I'm saying is that some userland stuff depends on CFQ | 10:17 |
| shadeslayer | and I was wondering if there was testing done to see if we can change to CFQ | 10:17 |
| shadeslayer | and wanted to look at testing data | 10:18 |
| shadeslayer | because I want to see cases where deadline is better on the desktop | 10:18 |
| shadeslayer | and see how kubuntu would be affected | 10:18 |
| cking | there has been testing done over a wide range of kernels and I doubt we will switch for this release | 10:18 |
| shadeslayer | yep, that's reasonable since we're past FF | 10:19 |
| shadeslayer | but data :( | 10:19 |
| shadeslayer | where do I find the data :( | 10:19 |
| cking | shadeslayer, it's currently not published because I'm still working on a fully automated test rig | 10:20 |
| cking | and with so many schedulers, file system times, media and kernels it's taking a while to gather a full set of data across releases + kernels | 10:20 |
| shadeslayer | okay, do you have a document that has testing parameters and such? | 10:21 |
| shadeslayer | cking: I assume the tests for the desktop are only run for Unity? | 10:21 |
| cking | shadeslayer, nope, it's all command line driven with 20 or so fio tests | 10:22 |
| shadeslayer | okay | 10:22 |
| cking | http://kernel.ubuntu.com/git?p=cking/fs-test-proto.git;a=summary | 10:22 |
| cking | and it's PROTO type code | 10:22 |
| shadeslayer | ack | 10:22 |
| cking | i.e. it's bound to have bugs and it's still work in progress | 10:23 |
| apw | shadeslayer, perhaps there is something to be said to getting your "select scheduler" job into something common to all servers etc to have different defaults | 10:25 |
| cking | shadeslayer, when I get some free time I'll try and write a fio test that reproduces that issue in that bug report | 10:25 |
| shadeslayer | apw: possibly | 10:26 |
| apw | it almost wants to be a kernel thing, though we don't really have "one package" due to the "multiple kernels installed" .... hmmm | 10:27 |
| shadeslayer | I just picked it up from the debian docs on udev rules, so I'm not really in the best position to write a more complex rule | 10:27 |
| apw | it would be nice if one had /etc/default/linux which let us change "things" which are kernel side, hmmm | 10:30 |
| cking | apw, yeah, like defaults for types of usecases, e.g. webserver, database, mailserver, desktop HDD,desktop SSD, etc | 10:35 |
| shadeslayer | cking: apw btw do these tests also test things like apps contending for disk reads/writes ? | 10:51 |
| shadeslayer | because for the desktop that would be the most important thing | 10:51 |
| cking | shadeslayer, some of my fio test do simulate multiple reader writers in different i/o patterns | 10:51 |
| === zequence_ is now known as zequence | ||
| === willcooke is now known as willcooke|meetin | ||
| === dannf` is now known as dannf | ||
| === alai` is now known as alai | ||
| === willcooke|meetin is now known as willcooke | ||
| === cmagina_ is now known as cmagina_airport | ||
| === cmagina_airport is now known as cmagina_mobile | ||
| === jsalisbury changed the topic of #ubuntu-kernel to: Home: https://wiki.ubuntu.com/Kernel/ || Ubuntu Kernel Team Meeting - Tues September 9th, 2014 - 17:00 UTC || If you have a question just ask, and do wait around for an answer! If the question is should I file a bug for something, likely you can assume yes. | ||
| === willcooke_ is now known as willcooke | ||
| === utlemming is now known as utlemming_away | ||
| JFSTWO | !ops | 20:22 |
| ubot5 | Help! lamont, zul, T-Bone, mdz, or jdub | 20:22 |
| ubot2` | Help! lamont, zul, T-Bone, mdz, or jdub | 20:22 |
| JFSTWO | !ops | 20:22 |
| ubot2` | Help! lamont, zul, T-Bone, mdz, or jdub | 20:22 |
| ubot5 | Help! lamont, zul, T-Bone, mdz, or jdub | 20:22 |
| bjf | dannf, we've not turned the crank on a new trusty yet. i expect rtg will get those ahci xgene patches sorted. don't know what happened there. | 21:12 |
| dannf | bjf: ack, thx | 21:12 |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!