Alternative Performance Kernel for Debian

[AdrianTM]

Leaving aside the douchebaggery, I think it would be interesting to come up with a clear(er) measurement for latency. Maybe launching hundreds of instances of a program at a time and still be able to do some useful work? Does anybody know how latency and responsiveness can be measured in an objective way?

[gnudude]

I haven't found any way to 'see' better performance from the liquorix kernels......maybe I am a douchebag...maybe...

[Telemachus]

I haven't found any way to 'see' better performance from the liquorix kernels......maybe I am a douchebag...maybe...

First, these are clearly two separate questions.

More seriously, I think that in ordinary use the average user would never see signficant effects from kernel tweaking. If you do video editing, heavy gaming or serious sound work, maybe. Otherwise, probably not. This has been my experience, at least, and I've experimented with many different kernels and patches over a number of years.

To be honest, the two things that I found I really feel - performance wise - are noatime and a faster spinning hard drive.

[gnudude]

I think I shall have to agree, not about the douchebag part but the performance part.

I guess I will have to leave the performance testing to the big boys and just stick to my noatime and fiddle with the different schedulers and filesystems...

[damentz]

For anyone interested, there is a BFS folder in http://liquorix.net/test. Try it out and see if anything breaks....

If not, I'll be using BFS in my kernel instead due to it's more reliable deadlines and average lower latency / higher throughput.

[damentz]

BFS v300

Code: Select all

real	0m56.723s
user	1m31.579s
sys	0m10.000s

The code in place to work around a certain bug ineffectively was removed, reducing the overhead. New kernels will be uploaded as 1.dmz.3a in the /test/ folder again.

EDIT: using BFS v300, not v240-test2. This version was simply renamed as there were no bugs.

[Conrad]

Hi,
your kernel '2.6.31-1.dmz.2-liquorix-amd64' works very well. I use Nvidia drivers, an X-Fi and Pulseaudio with some LADSPA effects. The various audio streams report latencies from 0 to about 0.5ms.

[damentz]

Hi,
your kernel '2.6.31-1.dmz.2-liquorix-amd64' works very well. I use Nvidia drivers, an X-Fi and Pulseaudio with some LADSPA effects. The various audio streams report latencies from 0 to about 0.5ms.

Thanks

My newer kernels are now using BFS, so you should achieve even less latency that you observed previously.

[Conrad]

Yes, now it's about 0.2msecs.

[tjwallace]

Trying this out now. Don't have any benchmarks but the overall system is now very snappy.

[alpha-dog]

liquorix has the 2.6.32 kernel out already. I installed it yesterday and the ATI proprietary drivers loaded without a problem.

[nadir]

Code: Select all

root@desktop-sid# cat /etc/apt/sources.list.d/liquorix.list 
deb http://liquorix.net/debian sid main

W: Failed to fetch http://liquorix.net/debian/dists/sid/ma ... ackages.gz Hash Sum mismatch

E: Some index files failed to download, they have been ignored, or old ones used instead.

how come? i had the same line in /etc/apt/sources.list.d/liquorix.list used successfully in the past. i had it commented out for a while, and am getting the error now.

[tjwallace]

Code: Select all

root@desktop-sid# cat /etc/apt/sources.list.d/liquorix.list 
deb http://liquorix.net/debian sid main

W: Failed to fetch http://liquorix.net/debian/dists/sid/ma ... ackages.gz Hash Sum mismatch

E: Some index files failed to download, they have been ignored, or old ones used instead.

how come? i had the same line in /etc/apt/sources.list.d/liquorix.list used successfully in the past. i had it commented out for a while, and am getting the error now.

Hmm I am om amd64 and getting the same thing...

W: Failed to fetch http://liquorix.net/debian/dists/unstab ... ackages.gz Hash Sum mismatch

E: Some index files failed to download, they have been ignored, or old ones used instead.

[nadir]

have you tried again? ( i tried several times and it didna work, if it wont soon i will forget )

in other words: if it will work, please post over here or pm me, so i will remember
tia

[craigevil]

# Liquorix sources added by smxi
deb http://liquorix.net/debian/ sid main

Works just fine here but my install is 32bit. No issues, kernel just updated a couple of days ago.

$ uname -a
Linux craigevil 2.6.31-5.dmz.3-liquorix-686 #1 ZEN SMP PREEMPT Mon Nov 2 05:27:44 UTC 2009 i686 GNU/Linux

[nadir]

# Liquorix sources added by smxi
deb http://liquorix.net/debian/ sid main

Works just fine here but my install is 32bit. No issues, kernel just updated a couple of days ago.

thanks for posting. i tried again and this time without problems. Might have been a temporary issue.

[damentz]

Sorry guys, the 2.6.32 kernel was a fluke! You can get it from the future repo: deb http://liquorix.net/debian sid main future

I accidentally threw it into the main one without knowing. This happened around the same time that the hash sum mismatch occurred as I was trying to restore my repository.

Everything is fixed now though, if you want to stay on 2.6.32, add the future repo to continue updating it! Otherwise you can remove the 2.6.32 packages and update to the latest 2.6.31 ones (2.6.31-5.dmz.4 as of this post).

PS: I'm very surprised that fglrx works. Preemptable RCU is known to cause problems with this driver module, but I'm using Preemptable Tree RCU now-a-days going forward with 2.6.32. It's supposed to be simpler and less buggy than the original Preemptable RCU implementation and so far I've had zero problems with it.

[alpha-dog]

PS: I'm very surprised that fglrx works. Preemptable RCU is known to cause problems with this driver module, but I'm using Preemptable Tree RCU now-a-days going forward with 2.6.32. It's supposed to be simpler and less buggy than the original Preemptable RCU implementation and so far I've had zero problems with it.

I used the sgfxi script -> this command

Code: Select all

sgfxi -f -! 1

[damentz]

Excellent, well I'm thrilled that fglrx can now use some sort of preemptable RCU, everyone wins here

[alpha-dog]

Excellent, well I'm thrilled that fglrx can now use some sort of preemptable RCU, everyone wins here

I've been useing 2.6.32-rc8 liquorix kernel and it works good, but it will freeze for up to a minute occasionally. I have read that the fglrx drivers will cause that, but I like the 3d and have kept it installed. I've also seen that the 2.6.32 kernel may not need the fglrx drivers for 3d. Will 3d work on the r6xx chipset ( radeon 3870 ) with open-source drivers with your 2.6.32-rc8 liquorix kernel. I also have a problem with shutdown, it will get to KILL ALL and hang.