View previous topic :: View next topic |
Author |
Message |
audiodef Watchman
Joined: 06 Jul 2005 Posts: 6639 Location: The soundosphere
|
Posted: Fri Nov 09, 2012 5:55 pm Post subject: |
|
|
I will try those settings on 3.6.2. 3.5.7-r1 seems to be working with the same config. I'm running JACK and will leave it running for a while as a test. If there are no xruns while I'm doing a bunch of stuff on the machine, WIN!
The only thing I notice, using 3.5.7-r1, is that switching to this machine on my hardware KVM switch takes longer than it does with any other kernel, bar none. I have to wait a few seconds for the keyboard to start working. It's almost instantaneous with other kernels using the exact same config. _________________ decibel Linux: https://decibellinux.org
Github: https://github.com/Gentoo-Music-and-Audio-Technology
Facebook: https://www.facebook.com/decibellinux
Discord: https://discord.gg/73XV24dNPN
Last edited by audiodef on Fri Nov 09, 2012 6:24 pm; edited 2 times in total |
|
Back to top |
|
|
audiodef Watchman
Joined: 06 Jul 2005 Posts: 6639 Location: The soundosphere
|
|
Back to top |
|
|
aCOSwt Bodhisattva
Joined: 19 Oct 2007 Posts: 2537 Location: Hilbert space
|
Posted: Fri Nov 09, 2012 6:28 pm Post subject: |
|
|
audiodef wrote: | Hey, what about timer frequency? I notice the options go up to 10k. |
Yes! ck-sources offer you the possibility to run over 1000 hz
Keep it quitely at 1000 unless you want to save power, in which case, drop down to 300.
Note the comments made about these extra options :
- "1500 Hz is an insane value"...
- "5000 Hz is an obscene value"...
And of course : "Being over 1000, driver breakage is likely." _________________
|
|
Back to top |
|
|
aCOSwt Bodhisattva
Joined: 19 Oct 2007 Posts: 2537 Location: Hilbert space
|
Posted: Fri Nov 09, 2012 6:31 pm Post subject: |
|
|
audiodef wrote: | The only thing I notice, using 3.5.7-r1, is that switching to this machine on my hardware KVM switch takes longer than it does with any other kernel, bar none. I have to wait a few seconds for the keyboard to start working. It's almost instantaneous with other kernels using the exact same config. |
Please note as stated in the OP that if you boot on KVM with 3.6.2 you *want* to set the kvm useflag. _________________
|
|
Back to top |
|
|
audiodef Watchman
Joined: 06 Jul 2005 Posts: 6639 Location: The soundosphere
|
Posted: Fri Nov 09, 2012 7:04 pm Post subject: |
|
|
Yeah, the acronym KVM gets confused a lot... sorry.
I have a little box to switch my keyboard and mouse between two computers. Not Kernel Virtual Machine.
Still a lag of seconds when switching back to this machine. What could cause that? Must be something in the patches specific to ck-sources.
I just made and tested those changes you suggested earlier. I'm running with the freq at 10kHz. Nothing bad has happened yet. What are the advantages of running at speeds greater than 1000? 1000 Needs to be the minimum for audio production, wondering why there are faster speeds here.
3.6.2 is a no-go. I can't compile nvidia-drivers against it and thus, no xorg. 3.5.7-r1 is working, though, and JACK starts up right quick now that I have the correct preemption model set. _________________ decibel Linux: https://decibellinux.org
Github: https://github.com/Gentoo-Music-and-Audio-Technology
Facebook: https://www.facebook.com/decibellinux
Discord: https://discord.gg/73XV24dNPN |
|
Back to top |
|
|
PaulBredbury Watchman
Joined: 14 Jul 2005 Posts: 7310
|
Posted: Fri Nov 09, 2012 9:13 pm Post subject: |
|
|
audiodef wrote: | wondering why there are faster speeds here. |
Code: | grep -A2 10000 /usr/src/linux/kernel/Kconfig.hz |
Quote: | 10000 Hz is an obscene value to use to run broken software that is Hz limited. |
I've never heard of any examples of such broken software. |
|
Back to top |
|
|
audiodef Watchman
Joined: 06 Jul 2005 Posts: 6639 Location: The soundosphere
|
|
Back to top |
|
|
aCOSwt Bodhisattva
Joined: 19 Oct 2007 Posts: 2537 Location: Hilbert space
|
Posted: Tue Nov 13, 2012 1:38 pm Post subject: |
|
|
audiodef wrote: | So there's really no actual advantage to setting the frequency to 1500 or 5000 Hz, even for audio production? |
Hrmmm... the short answer is : There is really no actual (or is it actually no real...) advantage to setting the frequency to such high values for who knows what he his doing and controls what he his doing accordingly.
If we keep RT-processes apart (they are managed specifically by the scheduler) and pay attention to more ordinary processes which can be either IO-bound (heavy use of I/O devices and most of the time spent waiting for I/O operations to complete) or CPU-bound (requiring a lot of CPU time).
The scheduler, which constantly monitors the cpu spent by the tasks, will quickly penalize the CPU-bound tasks by automatically lowering their priority.
Therefore I/O-bound tasks will quickly preempt the batch processes, no matter how long the time-slice is.
However, if you do not pay any particular attention to scheduling models or priorities you can fall into a trap : The scheduler does not know a-priori if the process you launch is a batch process or an interactive process. It needs some times to discover this by itself...
Which means that if you launch concurrently one CPU-bound process and one interactive process, the interactive process might have to wait for a whole time-slice before starting its execution.
Of course, processes which are sometimes CPU-bound, some other times I/O-bound will delay other processes similarly each time they change their behavior from io/bound to cpu crunchers.
As a consequence of this, if you don't want to care with RT-PRIOS and scheduling models && emerge -e world BUT imperatively need whatever program to be started in less than a millisecond... then you should consider increasing the timer frequency over 1000Hz. (Of course, because of task switching, the overhead will be increased)
Of course, wise users will first help the scheduler by increasing its knowledge a-priori, that is to say that they will nice -n 19 their makes, run SCHED_IDLEPRIO what they do not really care about and SCHED_ISO their favourite audio apps. _________________
|
|
Back to top |
|
|
ulenrich Veteran
Joined: 10 Oct 2010 Posts: 1480
|
Posted: Tue Nov 13, 2012 3:37 pm Post subject: |
|
|
Yes, but
Quote: | then you should consider increasing the timer frequency over 1000Hz. (Of course, because of task switching, the overhead will be increased) |
because of this overhead I decreased to 300 Hz and disabled NO_HZ. Think of 300 times in one single second the cpu gets a heavy IRQ event to reschedule (therefore stoping all other work). This is enough!
Though I don't know if you in favor of less power consumption -mobile reason- enable NO_HZ ... |
|
Back to top |
|
|
audiodef Watchman
Joined: 06 Jul 2005 Posts: 6639 Location: The soundosphere
|
Posted: Tue Nov 13, 2012 3:39 pm Post subject: |
|
|
I think I understand that. Linux audio systems are supposed to have programs started by certain groups, such as the realtime and audio groups, given the correct scheduling priority anyway via limits.conf. So it's not so important that a program start quickly so much as it is given priority. I guess that would mean 1kHz should be sufficient and higher frequencies might be overkill for audio production.
Do you have any idea why ck-sources takes so long to handle keyboard-video-mouse switching? _________________ decibel Linux: https://decibellinux.org
Github: https://github.com/Gentoo-Music-and-Audio-Technology
Facebook: https://www.facebook.com/decibellinux
Discord: https://discord.gg/73XV24dNPN |
|
Back to top |
|
|
aCOSwt Bodhisattva
Joined: 19 Oct 2007 Posts: 2537 Location: Hilbert space
|
Posted: Tue Nov 13, 2012 8:54 pm Post subject: |
|
|
audiodef wrote: | Do you have any idea why ck-sources takes so long to handle keyboard-video-mouse switching? |
What do you mean ? any figures ? any step to reproduce ? Which apps running ? _________________
|
|
Back to top |
|
|
audiodef Watchman
Joined: 06 Jul 2005 Posts: 6639 Location: The soundosphere
|
Posted: Wed Nov 14, 2012 3:28 pm Post subject: |
|
|
You need a keyboard-video-mouse switch. No particular anything running. It just takes several seconds for the keyboard and mouse to be grabbed when I switch back to my DAW when it's using ck-sources. I figure there must be something different in the patches because this has happened with NO other kernel I've ever used. Since my switch is USB, I would assume it has something to do with USB in the kernel patches (not the config, which works just fine in other kernels I use). _________________ decibel Linux: https://decibellinux.org
Github: https://github.com/Gentoo-Music-and-Audio-Technology
Facebook: https://www.facebook.com/decibellinux
Discord: https://discord.gg/73XV24dNPN |
|
Back to top |
|
|
aCOSwt Bodhisattva
Joined: 19 Oct 2007 Posts: 2537 Location: Hilbert space
|
Posted: Wed Nov 14, 2012 5:37 pm Post subject: |
|
|
audiodef wrote: | You need a keyboard-video-mouse switch. No particular anything running. It just takes several seconds for the keyboard and mouse to be grabbed when I switch back to my DAW when it's using ck-sources. I figure there must be something different in the patches because this has happened with NO other kernel I've ever used. Since my switch is USB, I would assume it has something to do with USB in the kernel patches (not the config, which works just fine in other kernels I use). |
I don't get such device.
Do you get other troubles with other usb devices in general ?
Can you post the lines of your kernel log related to the attachment of this (I presume HID) device by its driver. Something looking like this :
Code: |
input Logitech HID compliant keyboard as /devices/pci0000:00/0000:00:1d.2/usb8/8-2/8-2:1.0/input/input3
hid-generic 0003 46D:C30E.0002: input: USB HID v1.10 Keyboard [Logitech HID compliant keyboard] on usb-0000:00:1d.2-2/input0
input Logitech HID compliant keyboard as /devices/pci0000:00/0000:00:1d.2/usb8/8-2/8-2:1.1/input/input4
hid-generic 0003 46D:C30E.0003: input: USB HID v1.10 Device [Logitech HID compliant keyboard] on usb-0000:00:1d.2-2/input1 |
On a side note, the .config you posted earlier says : CONFIG_USB_SUSPEND=y
Do you actually need that ? _________________
|
|
Back to top |
|
|
audiodef Watchman
Joined: 06 Jul 2005 Posts: 6639 Location: The soundosphere
|
|
Back to top |
|
|
aCOSwt Bodhisattva
Joined: 19 Oct 2007 Posts: 2537 Location: Hilbert space
|
Posted: Sat Nov 17, 2012 8:46 pm Post subject: |
|
|
Following Bug 443566 and thanks Markos Chandras, 3.4.18 is out as latest release of the long term stable branch. _________________
Last edited by aCOSwt on Thu Dec 20, 2012 10:23 am; edited 1 time in total |
|
Back to top |
|
|
audiodef Watchman
Joined: 06 Jul 2005 Posts: 6639 Location: The soundosphere
|
|
Back to top |
|
|
aCOSwt Bodhisattva
Joined: 19 Oct 2007 Posts: 2537 Location: Hilbert space
|
Posted: Thu Dec 13, 2012 7:30 pm Post subject: |
|
|
Following Bug 447008 - and thanks to Markos Chandras, 3.4.23 is out as latest release of the long term stable branch. _________________
Last edited by aCOSwt on Thu Dec 20, 2012 10:24 am; edited 1 time in total |
|
Back to top |
|
|
audiodef Watchman
Joined: 06 Jul 2005 Posts: 6639 Location: The soundosphere
|
|
Back to top |
|
|
aCOSwt Bodhisattva
Joined: 19 Oct 2007 Posts: 2537 Location: Hilbert space
|
Posted: Wed Dec 19, 2012 3:57 pm Post subject: |
|
|
audiodef wrote: | Earlier in this thread, I mentioned switching between two machines with a hardware KVM switch being slow when using ck-sources.
Well, it's clearly not the ck patches. It looks like something in the kernel itself, because I have the same problem when using the latest 3.6 rt-sources. Maybe I should file a kernel bug... |
I cannot tell if this might account for such an important delay but I have read somewhere an user reporting similar troubles that he had found directly linked with the dynticks feature.
If you built your system tickless, maybe you can try booting after appending nohz=off to your boot command line in order to test if things are better.
And, BTW, can you post the lines of your kernel log related to the attachment of the (I presume HID) device by its driver ? _________________
|
|
Back to top |
|
|
aCOSwt Bodhisattva
Joined: 19 Oct 2007 Posts: 2537 Location: Hilbert space
|
Posted: Wed Dec 19, 2012 7:35 pm Post subject: |
|
|
Following Bug 447718 - and thanks to Markos Chandras, 3.6.11 is out.
It will be the last release of the 3.6 branch and all users of 3.5.7 or 3.6.2 should upgrade. It is actually worth it. _________________
|
|
Back to top |
|
|
aCOSwt Bodhisattva
Joined: 19 Oct 2007 Posts: 2537 Location: Hilbert space
|
Posted: Sat Feb 23, 2013 9:49 pm Post subject: |
|
|
Following Bug 458874 - and thanks to Markos Chandras, 3.4.32 is out as latest release of the long term stable branch.
All 3.4 users should upgrade.
Per same bug, 3.7.9 is out too but... about it... no wise user, I mean no user using the ck-sources<3.7.9 for work, is invited to "upgrade".
BTW, with the 3.7.9 a patch is being applied on the "legacy" bfs-426, fixing a couple of bugs. (corner cases)
If those bugs are impacting on 3.4.x or 3.6.11 users, I'll make my best to port this patch to these releases.
Just tell. _________________
|
|
Back to top |
|
|
aCOSwt Bodhisattva
Joined: 19 Oct 2007 Posts: 2537 Location: Hilbert space
|
Posted: Sun Mar 03, 2013 6:26 pm Post subject: |
|
|
Following BUG 459926 and thanks to Markos Chandras, the ck-sources tree has been updated in order to fix the out-of-bounds access to sock_diag_handlers security issue as reported in BUG 459124
3.7 users should upgrade to 3.7.10,
3.6 users should upgrade to 3.6.11-r1,
3.4 users should upgrade to 3.4.34
The security fix has been backported to 3.5.7-r2 as well for the convenience of 3.5 users who cannot upgrade to 3.6 _________________
|
|
Back to top |
|
|
ulenrich Veteran
Joined: 10 Oct 2010 Posts: 1480
|
Posted: Mon Mar 04, 2013 1:24 pm Post subject: |
|
|
@aCOSwt, and there is a new Linux-3.8.2 Bfs which runs perfectly well:
http://ck-hack.blogspot.de/2013/03/bfs-0428-for-linux-38x.html
Con Kolivas not only has adapted the BrainFuckScheduler to Linux-3.8 but also made at least two corrections to time accounting. Which I previously grumbled about at his blog, because as a simple user not able to handle the perf infrastructure I feel dependend on such tools as top to audit my own system.
Con also erased two of his ck1 patches:
mm-decrease_default_dirty_ratio-1.patch
mm-minimal_swappiness.patch
Which he thinks should be handled by distributions charge of sysctl. Surely these patches were based on old assumptions not any more true, because of such newer things as HugePages, especially when running virtual machines.
I think applying Cons hz-default_1000.patch is good for gamers.
But pure office users and developers, who often call gcc to compile,
reach better throughput by options like: Code: | CONFIG_RCU_BOOST=y
CONFIG_RCU_BOOST_PRIO=88
# CONFIG_NO_HZ is not set
CONFIG_HZ_300=y
CONFIG_HZ=300 |
|
|
Back to top |
|
|
aCOSwt Bodhisattva
Joined: 19 Oct 2007 Posts: 2537 Location: Hilbert space
|
Posted: Mon Mar 04, 2013 4:06 pm Post subject: |
|
|
ulenrich wrote: | @aCOSwt, and there is a new Linux-3.8.2 Bfs which runs perfectly well: |
I am currently testing it and, if successful, I'll push a 3.8.3
ulenrich wrote: | Con Kolivas not only has adapted the BrainFuckScheduler to Linux-3.8 but also made at least two corrections to time accounting. |
You are correct but these corrections have been made in bfs-427.
Early 3.7 were patched with 426 but I ship the 3.7.10 of the ck-sources with 427. => The ck-sources-3.7.10 users already enjoy these corrections you mention.
ulenrich wrote: | Con also erased two of his ck1 patches:
mm-decrease_default_dirty_ratio-1.patch... Which he thinks should be handled by distributions charge of sysctl. Surely these patches were based on old assumptions not any more true, because of such newer things as HugePages, especially when running virtual machines. |
As for the dirty-ratios, not exactly.
Con removed the setting of the defaults following a discussion I had with him on IRC demonstrating that just "plenty" of programs were arbitrarily "playing" with the dirty-ratios from boot time, rendering vain the action of this particular patch. _________________
|
|
Back to top |
|
|
thunderrd n00b
Joined: 20 Aug 2010 Posts: 59
|
Posted: Tue Mar 05, 2013 4:15 pm Post subject: |
|
|
Eric, for what it's worth, I've been running the 3.8.2-ck as well and it looks quite fine from where I sit
I think your plan is good, test it for a week or so more, then release a 3.8* to everyone, with the bfs-427, which is really important. TBH, I skipped over the 3.7.x versions on my boxen waiting for the next major update. [I would think that many others did, as well.]
So I will be going directly from 3.6.11 to 3.8.x
Regards,
tr |
|
Back to top |
|
|
|