https://features.opensuse.org/310843

thanks,i see it is still unconfirmed :(. -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org

On 11/18/2010 6:39 AM, Cristian Rodríguez wrote:

El 18/11/10 00:01, michael getachew escribió:

...

Hello, i just tried out the 200 line kernel patch everyone is talking about and does it make a big difference or what? i can actually do a make -j64 on the linux kernel source code,compile firefox and browse the web seamlessly .

my question is will a patched kernel be found in opensuse repos anytime soon or are we stuck with manually patching/compiling kernels?

This patch has been enabled in Kernel:HEAD for the *-desktop variants.

Is the patch available for Debian-derived variants, and if so, what would it be called in, say, Symantic Package Manager? --doug -- Blessed are the peacemakers...for they shall be shot at from both sides. --A.M. Greeley -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org

On Thu, 18 Nov 2010 23:56:58 +0530, Cristian Rodríguez wrote:

the current kernel:HEAD will cause X not to start due to a bug, I guess it will be fixed soon.

i just installed 2.6.37-rc2-1-desktop, and it runs fine, incl. X. perhaps this is an extension of the recent nvidia driver problem, which affects some but not all? -- phani. -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org

El 18/11/10 16:21, Matt Hayes escribió:

Agreed. I'm running the latest Kernel:HEAD as well with the latest Nvidia driver, no issue here at all. Runs smoothly.

This bug doesnt trigger with the Nvidia propietary drivers, only with (ones Im using BTW..) it is this bug: http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=... -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org

If anybody's interested, there is aparently a userspace equivalent to the kernel patch: http://lkml.org/lkml/2010/11/16/392 I haven't tried it myself. Cheers, Dave -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org

El 20/11/10 04:45, Ian escribió:

On Friday 19 November 2010 13:18:48 Dave Howorth wrote:

...

If anybody's interested, there is aparently a userspace equivalent to the kernel patch:

http://lkml.org/lkml/2010/11/16/392

I haven't tried it myself.

I have but I can't get passed the mount option.

You need a kernel with CGROUPS enabled. -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org

Anton Aylward wrote:

Cristian Rodríguez said the following on 11/20/2010 07:49 AM:

...

El 20/11/10 04:45, Ian escribió:

...

On Friday 19 November 2010 13:18:48 Dave Howorth wrote:

...

If anybody's interested, there is aparently a userspace equivalent to the kernel patch:

http://lkml.org/lkml/2010/11/16/392

I haven't tried it myself.

I have but I can't get passed the mount option.

You need a kernel with CGROUPS enabled.

Well I installed 2.6.37-rc2-1 which has CGROUPS enabled and /sys/fs/cgroup but the mount fails:

# mount -t cgroup cgroup /sys/fs/cgroup/cpu -o cpu mount: mount point /sys/fs/cgroup/cpu does not exist

In fact # ls /sys/fs/cgroup/

i.e. nothing there. So "obviously" those instructions make some assumption tha deons't hold for us. Some initialization perhaps?

I have 2.6.36-18-pae, the mount point exists. I don't remember having to do anything extra. -- Per Jessen, Zürich (6.4°C) -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org

Per Jessen said the following on 11/20/2010 08:53 AM:

Per Jessen wrote:

...

I have 2.6.36-18-pae, the mount point exists. I don't remember having to do anything extra.

Ignore that, same issue here, i.e. /sys/fs/cgroup/cpu does not exist.

As far as I can see, there is supposed to be some initialization code in /etc/init.d/cgconfig that expect there to have been "cgroup"entries in /etc/fstab and hence /proc/mounts There isthen the 'rules engine' started in /etc/init.d/cgred Trying to start either of those fails. -- If I had never thought about computer typesetting, I might have had a happier life in some ways. -- Knuth -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org

Per Jessen said the following on 11/20/2010 12:23 PM:

Anton Aylward wrote:

...

Per Jessen said the following on 11/20/2010 08:53 AM:

...

Per Jessen wrote:

...

I have 2.6.36-18-pae, the mount point exists. I don't remember having to do anything extra.

Ignore that, same issue here, i.e. /sys/fs/cgroup/cpu does not exist.

As far as I can see, there is supposed to be some initialization code in /etc/init.d/cgconfig that expect there to have been "cgroup"entries in /etc/fstab and hence /proc/mounts

It looks like it's expecting a file /etc/cgconfig.conf first of all. There is a man page for that. I created one like this:

mount { cpu = /mnt/cgroups/cpu; }

Then I mounted:

mount -t cgroup cgroup /mnt/cgroups/cpu -o cpu

So far so so good, but the bashrc script still expects to find a 'user' directory under cpu/, which isn't there.

Try the "ubuntu" variant, with /etc/rc.d/boot.local, but with "/mnt" instead of "/dev" and that includes the line mkdir -m 0777 /mnt/cgroup/cpu/user You can reboot or jsut run that file by hand as root. The the ~/.bashrc for "unbuntu" with the mnt/dev tweak as well. Hmm. It ought to go into ~/.xinitrc as well ?? -- Though force can protect in emergency, only justice, fairness, consideration and co-operation can finally lead men to the dawn of eternal peace. -- Dwight D. Eisenhower -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org

Doug wrote:

On 11/20/2010 06:49 AM, Cristian Rodríguez wrote:

...

El 20/11/10 04:45, Ian escribió:

...

On Friday 19 November 2010 13:18:48 Dave Howorth wrote:

...

If anybody's interested, there is aparently a userspace equivalent to the kernel patch:

http://lkml.org/lkml/2010/11/16/392

I haven't tried it myself. I have but I can't get passed the mount option.

You need a kernel with CGROUPS enabled. How would you know if your kernel has CGROUPS enabled?

zgrep CGROUP /proc/config.gz -- Per Jessen, Zürich (3.5°C) -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org

On 20.11.2010, Per Jessen wrote:

...

You really own a machine with 64 processors/threads? If not, the parameter you pass to "make" with the "-j" option should not exceed the number of cpus present.

Why not? - if work is somewhat IO-bound, running more jobs than available processors is quite reasonable.

He's doing a "make -j64" and his machine is going to freeze. Unless there's proof that his machine is capable of this large amount of processes, I'm arguing that this is one of the main causes which gets the machine down to the knees. I would at least have checked with "vmstat 1" and "iostat" while running "make -j64"... -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 On Sunday, 2010-11-21 at 10:47 +0100, Per Jessen wrote:

It's not the CPU-load but lack of memory that'll stop it though.

Then, what it needs is an option that instead of launching a fixed number of threads, launchs a variable number of threads that doesn't cause swap to be used. Dynamically checking overall system load. - -- Cheers, Carlos E. R. (from 11.2 x86_64 "Emerald" at Telcontar) -----BEGIN PGP SIGNATURE----- Version: GnuPG v2.0.12 (GNU/Linux) iEYEARECAAYFAkzpIs8ACgkQtTMYHG2NR9WfTgCZAX0sYLIP1qdOdj3Dz06k+hV7 4nYAoJkjgRPBgkbIkxdD20hvOrTGfSLc =s6ha -----END PGP SIGNATURE----- -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org

On 11/21/2010 9:05 AM, Per Jessen wrote:

Carlos E. R. wrote:

...

On Sunday, 2010-11-21 at 10:47 +0100, Per Jessen wrote:

...

It's not the CPU-load but lack of memory that'll stop it though.

Then, what it needs is an option that instead of launching a fixed number of threads, launchs a variable number of threads that doesn't cause swap to be used. Dynamically checking overall system load.

If anything else uses swap, it'll never start. Besides, it's not really a job for 'make' (and others who submit jobs in parallel, e.g. xargs).

Did you guys miss the point that he only used -j64 to try to load the system in a way that would normally result in terrible user interactive desktop perfromance, and test that the new cgroups hack causes the desktop and other user interactive processes to get cpu and i/o scheduling no matter how loaded the rest of the system is? A method that dynamically runs the correct optimal -jN is exactly what you don't want in a stress test. -- bkw -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org

No, I didn't. If you start swaping, the cpu stress point is missed, because you get into the disk i/o stress point. The system will become unresponsive for a different reason and you can't see if the new scheduler is working slow because the patch doesn't work or because it is waiting for the disk.

i agree i did the same -j64 compile and compile firefox too at the same time and the system was very responsive. until it eventually ended up freezing the system. and i still see my system get unresponsive under stress at times like if i was running 2 virtual machines and i only have <100Mb memory left,etc... this is not a patch that will make your system responsive in any condition. but it does make it responsive when the cpu is under lots of stress. this is specially useful for multi-threaded applications like if you were encoding video and playing your favorite 3d game and downloading a bunch of videos/software at the same time(provided you have adequate memory/swap). i think it is mainly targeted for desktop responsiveness but i can imagine how it might be useful on servers with high-load . -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org

Carlos E. R. wrote:

On Sunday, 2010-11-21 at 15:05 +0100, Per Jessen wrote:

...

Carlos E. R. wrote:

...

On Sunday, 2010-11-21 at 10:47 +0100, Per Jessen wrote:

...

It's not the CPU-load but lack of memory that'll stop it though.

Then, what it needs is an option that instead of launching a fixed number of threads, launchs a variable number of threads that doesn't cause swap to be used. Dynamically checking overall system load.

If anything else uses swap, it'll never start. Besides, it's not really a job for 'make' (and others who submit jobs in parallel, e.g. xargs).

I think that yes, it is the task. There are other methods instead of checking directly swap: check if your own program is hitting swap, perhaps. If it is, the advantage of launching more threads is lost.

We're going way OT, so just a final comment - I definitely don't think any of that is a job for make nor any other utility that can do stuff in parallel. It is the responsibility of the system to try to load the system optimally. /Per -- Per Jessen, Zürich (4.0°C) -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org