Re: Low Latency vs. Real Time Kernel - actual latencies ?

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Here's a diff of between the configs of Ubuntu's generic and lowlatency
linux-image (amd64) packages:

$ diff config-generic config-lowlatency
3c3
< # Linux/x86_64 3.19.0-12-generic Kernel Configuration
---
> # Linux/x86_64 3.19.0-12-lowlatency Kernel Configuration
69c69
< CONFIG_VERSION_SIGNATURE="Ubuntu 3.19.0-12.12-generic 3.19.3"
---
> CONFIG_VERSION_SIGNATURE="Ubuntu 3.19.0-12.12-lowlatency 3.19.3"
96c96
< # CONFIG_IRQ_FORCED_THREADING_DEFAULT is not set
---
> CONFIG_IRQ_FORCED_THREADING_DEFAULT=y
135c135
< CONFIG_TREE_RCU=y
---
> CONFIG_PREEMPT_RCU=y
145a146
> # CONFIG_RCU_BOOST is not set
251d251
< CONFIG_OPTPROBES=y
381,385d380
< CONFIG_INLINE_SPIN_UNLOCK_IRQ=y
< CONFIG_INLINE_READ_UNLOCK=y
< CONFIG_INLINE_READ_UNLOCK_IRQ=y
< CONFIG_INLINE_WRITE_UNLOCK=y
< CONFIG_INLINE_WRITE_UNLOCK_IRQ=y
457,458c452,454
< CONFIG_PREEMPT_VOLUNTARY=y
< # CONFIG_PREEMPT is not set
---
> # CONFIG_PREEMPT_VOLUNTARY is not set
> CONFIG_PREEMPT=y
> CONFIG_PREEMPT_COUNT=y
565c561
< CONFIG_HZ_250=y
---                                                                                                                                                         

> # CONFIG_HZ_250 is not
set                                                                                                                                

567,568c563,564                                                                                                                                             

< # CONFIG_HZ_1000 is not
set                                                                                                                               

< CONFIG_HZ=250
---
> CONFIG_HZ_1000=y
> CONFIG_HZ=1000
4998d4993
< CONFIG_DRM_I810=m
7487a7483
> # CONFIG_DEBUG_PREEMPT is not set
7551a7548
> # CONFIG_PREEMPT_TRACER is not set

So it looks like the big differences with the lowlatency config are
enabling CONFIG_PREMPT, CONFIG_IRQ_FORCED_THREADING_DEFAULT, and
CONFIG_PREEMPT_RCU as well as setting CONFIG_HZ to 1000. Would using
these settings in a Fedora kernel have the stability concerns as the RT
patch set? Would there be any other drawbacks?

On 04/05/2015 12:21 PM, Be wrote:
> Has anyone looked into what Ubuntu Studio is doing with the lowlatency
> kernel? Would it be feasible to include a similarly configured kernel in
> Fedora?
>
> On 11/13/2014 10:59 AM, Brian Monroe wrote:
>> I think so too, thanks for chiming in. 
>>
>> I'm still waiting to get into the packagers group, but I have a koji
>> account and theoretically could compile an rt kernel. I think the
>> standard naming schema in other distros is kernel-rt. It should be
>> only adding a few lines to the spec file to enable the rt kernel, but
>> when you look at how many kernel update there are for Fedora every
>> week, I'm not sure as to how up to date we'll be able to keep up due
>> to the work load. We're already are down on developers, and people
>> like Brandon are keeping us afloat.
>>
>> Are we going to be ok as a project to be behind a week or two in
>> Kernel releases? Personally I'm for more stable kernels when it comes
>> to music production vs. having the latest and greatest, but I also
>> think that should be a clearly indicated as a feature 
>>
>> That being said, I feel strongly as though others should take this
>> task on, if not me, then someone else or better yet, a few of us. 
>>
>>
>> I'm looking into the Ubuntu Studio and turns out they dropped the RT
>> kernel as default. They're using a "lowlatency" kernel instead of a rt
>> kernel (though they do still supply an rt kernel but not by default).
>> I do know that users are able to get 1.5 ms latency with zero xruns so
>> I'm guessing they're doing something other than real-time scheduling,
>> I just don't know what. Thoughts?
>>
>> On Wed Nov 12 2014 at 10:40:44 AM Be Ing <be.0@xxxxxxx
>> <mailto:be.0@xxxxxxx>> wrote:
>>
>>     Hello Fedora musicians, I've been lurking this list for a little
>>     bit and this is my first time chiming in on something.
>>
>>     I think it is important to pursue an official realtime kernel for
>>     Fedora. I think a distribution focused on audio without a realtime
>>     kernel would have a serious bug, that IMO, would be worth delaying
>>     publication for.
>>
>>     >So I had a beer with hansomepirate(jdulaney), who is, or was on
>>     the kernel
>>     sig, last night and we got to talking about a RT kernel.
>>     >
>>     >Last time we talked to the kernel folks about an rt kernel, they
>>     weren't
>>     impressed with the "need" for Fedora, but that was before the Spin was
>>     officially out.
>>     >
>>     >Now might be a good time to raise this issue again? I dug through my
>>     archives and found this thread. Now that we have an actual spin
>>     that's out,
>>     we can actually redo some of the testing to have more realistic tests.
>>     (multitrack with effects)
>>     >
>>     >I feel like right now, it's one of the few benefits that the
>>     ubuntu studio
>>     folks have (or at least claim to have) over us. The other is some
>>     semi-proprietary software that on... you know what, never mind
>>     it's getting
>>     off topic.
>>     >
>>     >Anyways, does the list think this is worth pursuing?
>>     >
>>     >>On Wed Feb 22 2012 at 9:10:29 PM Brian Monroe <briancmonroe at
>>     gmail.com
>>     <http://gmail.com>[https://admin.fedoraproject.org/mailman/listinfo/music]>
>>     wrote:
>>     >>
>>     >> Ok, I redid all the tests, while the system was only running my
>>     DE and the
>>     >> test, and then again when I put it under duress by running a
>>     script that
>>     >> looped "du -h /" and "ls -Ral /usr/" over and over. I ran the
>>     script twice
>>     >> to get my proc up a bit to emulate running some intese delays
>>     and reverbs
>>     >> or other effects.
>>     >>
>>     >> Ironically the kernels typically did better when the scripts
>>     were running.
>>     >> Personally I think there's a clear advantage with CCRMA's
>>     kernel or even
>>     >> just a preempt kernel in the max lat areas. Those max numbers
>>     jumped up
>>     >> close to where they were near the beggining of the test if
>>     anyone was
>>     >> wondering.
>>     >>
>>     >> Here's the file with both sets of tests and the uname -a info
>>     as requested
>>     >> by Fernando.
>>     >> -Brian
>>     >>
>>     >>> On Wed, Feb 22, 2012 at 6:54 AM, Brian Monroe <briancmonroe at
>>     gmail.com
>>     <http://gmail.com>[https://admin.fedoraproject.org/mailman/listinfo/music]
>>     >>> wrote:
>>     >>> I'll be sure to include that on the next batch. I used the
>>     kernel you
>>     >>> after installing the CCRMA repo when you use yum install
>>     kernel-rt, which
>>     >>> happens to be 3.0.17-1.rt33.1.fc16.ccrma.x86_64.rt. I'll go
>>     back and
>>     >>> include the other info to the old results when I do the load
>>     testing
>>     >>> tonight or tomorrow.
>>     _______________________________________________
>>     music mailing list
>>     music@xxxxxxxxxxxxxxxxxxxxxxx <mailto:music@xxxxxxxxxxxxxxxxxxxxxxx>
>>     https://admin.fedoraproject.org/mailman/listinfo/music
>>
> _______________________________________________
> music mailing list
> music@xxxxxxxxxxxxxxxxxxxxxxx
> https://admin.fedoraproject.org/mailman/listinfo/music

_______________________________________________
music mailing list
music@xxxxxxxxxxxxxxxxxxxxxxx
https://admin.fedoraproject.org/mailman/listinfo/music





[Index of Archives]     [Linux Audio Users]     [ALSA Users]     [Fedora Development]     [Fedora Desktop]     [Fedora Users]     [Gimp]     [Yosemite News]

  Powered by Linux