| OvenWerks | (simplify things) | 00:00 |
|---|---|---|
| OvenWerks | Actually this seems to be only three files: | 00:08 |
| OvenWerks | /etc/grub.d/09_lowlatency Makes Lowlatency kernel default | 00:09 |
| OvenWerks | /etc/sysctl.d/50-ubuntustudio.conf sets swappiness = 10 and has some commented out settings we could not confirm. | 00:10 |
| OvenWerks | /lib/udev/rules.d/40-timer-permissions.rules Allows user with audio to set grain on rtc0 and hpet. I think I have heard that "properly" coded applications don't need it... but there are bound to be some that still do. | 00:13 |
| OvenWerks | This last one needs to be remembered if jackd starts getting RT permisions some other way than the audio group. | 00:14 |
| OvenWerks | zequence: instead of a new package, the first one should really be a part of the lowatency kernel... but that would probably mean we would have to maintain it :P | 00:16 |
| OvenWerks | The other two could be part of -audio-core | 00:17 |
| OvenWerks | Though to be honest, I would like to see swappiness = 10 upstreamed as default for generic as well because any feed back I have gotten has been this is good for all desktop work. | 00:19 |
| OvenWerks | However the truth is I probably don't know enough about what I am talking about. | 00:40 |
| OvenWerks | Thereason swappiness matters in audio is a little more complex. Many audio applications lock memory so it can't be swapped and are not a problem. However, some programs that don't do audio themselves, but still have jack ports open for monitoring and config reasons don't (qjackctl for example). If the GUI of these programs gets swapped and a button is clicked, that jack port gets frozen long enough to cause audible xruns. In my opinion this is re | 00:46 |
| OvenWerks | It seems I found even the pager being swapped out caused problems if I needed to switch from workspace to another. | 00:49 |
| OvenWerks | zequence: Actually, those three files could be moved to -controls itself if we load it by default. | 00:55 |
| OvenWerks | zequence: https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/plain/Documentation/scheduler/sched-deadline.txt?id=712e5e34aef449ab680b35c0d9016f59b0a4494c | 01:23 |
| OvenWerks | ^^^ this is new since kernel 3.14. Can we use this to improve lowlatency? | 01:24 |
| OvenWerks | swappiness==0 is handled in a unique way from all other values: https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=8582cb96b0bfd6891766d8c30d759bf21aad3b4d | 01:33 |
| OvenWerks | Is there a reason we do not include jack.tools? | 06:24 |
| zequence | OvenWerks: We can worry about uploads later, once we get going with this cycle development. I'm trying to get upload rights, so hoping that will sort itself out within the week or so | 08:30 |
| zequence | You bring up a lot of things there. Let me get started, and I'll comment later | 08:30 |
| OvenWerks | zequence_: no problem, I am not rushing anything, just doing some research. From reading though some things, I would say there is zero chance of changing swappiness by default upstream. It has already been discussed to death upstream and not changed. | 13:02 |
| zequence_ | OvenWerks: only a matter of time until rt privilege is solved http://trac.jackaudio.org/wiki/Cgroups | 13:10 |
| === zequence_ is now known as zequence | ||
| zequence | But still, until it is fixed, we should make rt administration easy. On trusty, specifically | 13:12 |
| zequence | We do a SRU to make -controls at all usable on trusty. A more enhanced version, we either backport or do a SRU - depending on if we are allowed to SRU, or not | 13:13 |
| zequence | By the time of 16.04 I would assume things have changed | 13:13 |
| zequence | ..so we can drop realtime privilege administration | 13:14 |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!