[gentoo-amd64] compiling on dual-cores - only 50% usage per cpu

[gentoo-amd64] compiling on dual-cores - only 50% usage per cpu

[Bernhard Auzinger]

Hi everybody,

it's not a important question, but has anybody of you noticed when compiling 
certain packages the load on both cpu's is only about 50%. I have the feeling 
that it happens with packages that can not be split up into two jobs. But in 
this case one cpu should be on 100% load and the other one at ~0%. Maybe 
someone of you has a good explanation for this behaviour.

Rgds
Bernhard
-- 
gentoo-amd64@gentoo.org mailing list

Re: [gentoo-amd64] compiling on dual-cores - only 50% usage per cpu

[Kris Kersey (Augustus)]

You have it right.  If a package is not able to build in parallel or if 
the package exhibits problems because of parallel builds and has had the 
make flags disabled, then you would only see one file compiled at a time. 
This will sometimes load one processor at 100% and other times, it will 
load each processor a fraction of 100%.  Keep in mind that the processor 
statistics you're looking at is over a certain period of time (say one 
second) and therefor, over that period, it averaged a number smaller than 
100%.

Thanks,
Kris Kersey (Augustus)
LinuxHardware.org Site Manager
augustus@linuxhardware.org
Gentoo Linux AMD64 Developer
augustus@gentoo.org
AIM: Augustus22

On Sat, 10 Nov 2007, Bernhard Auzinger wrote:

> Hi everybody,
>
> it's not a important question, but has anybody of you noticed when compiling
> certain packages the load on both cpu's is only about 50%. I have the feeling
> that it happens with packages that can not be split up into two jobs. But in
> this case one cpu should be on 100% load and the other one at ~0%. Maybe
> someone of you has a good explanation for this behaviour.
>
> Rgds
> Bernhard
> -- 
> gentoo-amd64@gentoo.org mailing list
>
>
-- 
gentoo-amd64@gentoo.org mailing list

Re: [gentoo-amd64] compiling on dual-cores - only 50% usage per cpu

[Conway S. Smith]

[-- Attachment #1: Type: text/plain, Size: 854 bytes --]

On Sat, 10 Nov 2007 17:29:24 +0100
Bernhard Auzinger <e0026053@student.tuwien.ac.at> wrote:
> Hi everybody,
> 
> it's not a important question, but has anybody of you noticed when
> compiling certain packages the load on both cpu's is only about 50%.
> I have the feeling that it happens with packages that can not be
> split up into two jobs. But in this case one cpu should be on 100%
> load and the other one at ~0%. Maybe someone of you has a good
> explanation for this behaviour.
> 
> Rgds
> Bernhard

If you have a strong preference to having a process only use one CPU,
check out schedutils & CPU affinity.  I also seem to recall reading
about a sysctl that let you modify how willing the scheduler is to
switch CPUs, but my quick google just now didn't find that, so maybe
I'm making that up.


Good luck,
Conway S. Smith

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-amd64] compiling on dual-cores - only 50% usage per cpu

[Ângelo Miguel Arrifano]

On Sat, 10 Nov 2007 17:29:24 +0100
Bernhard Auzinger <e0026053@student.tuwien.ac.at> wrote:

> Hi everybody,
> 
> it's not a important question, but has anybody of you noticed when compiling 
> certain packages the load on both cpu's is only about 50%. I have the feeling 
> that it happens with packages that can not be split up into two jobs. But in 
> this case one cpu should be on 100% load and the other one at ~0%. Maybe 
> someone of you has a good explanation for this behaviour.
> 
> Rgds
> Bernhard
> -- 
> gentoo-amd64@gentoo.org mailing list
> 

I made a core metering plugin for XFCE panel some time ago.
If you use XFCE or are interested on the plugin somehow, you can get it here:

http://miknix.homelinux.com/xfce4-cpuload-plugin/

Looking to the plugin it is very visible what you say about the overall when processor
usage is 50% one of the cores is 100% and the other at 0%.
I can also state that most of single processes running on a single core, are always
migrating from one core to another maybe to distribute temperature over both cores.

The exception (at least on my hardware) is when some process is making exhaustive
I/O using interrupts. Both cores are at 0% usage but the overall cpu usage is 100%.

Also, there are packages that fails using parallel make. These packages are overriding
your MAKEOPTS="-jx" option with -j1. But, if you find that some package is overriding
the flag but builds OK with -j>1, please report on bugzilla! Specially if is a big package :) 

Best regards,
-- 
Angelo Arrifano AKA MiKNiX
CSE Student at UBI, Portugal
Gentoo Linux AMD64 Arch Tester
miknix@gmail.com
http://miknix.homelinux.com
PGP Pubkey online

- -
He is a man capable of turning any colour into grey.
		-- John LeCarre 
-- 
gentoo-amd64@gentoo.org mailing list

Re: [gentoo-amd64] compiling on dual-cores - only 50% usage per cpu

[Bernhard Auzinger]

Thank you for your explanations. It seems to make sense that the job hops from 
one CPU to another, just too fast to be seen on the averaged processor 
statistics. I think I'll take a look at schedutils. The xfce-cpuload-plugin 
seems to be nice, but I'm kde-user :).

Rgds
Bernhard
-- 
gentoo-amd64@gentoo.org mailing list

[gentoo-amd64] Re: compiling on dual-cores - only 50% usage per cpu

[Duncan]

Bernhard Auzinger <e0026053@student.tuwien.ac.at> posted
200711102143.03343.e0026053@student.tuwien.ac.at, excerpted below, on 
Sat, 10 Nov 2007 21:43:03 +0100:

> Thank you for your explanations. It seems to make sense that the job
> hops from one CPU to another, just too fast to be seen on the averaged
> processor statistics. I think I'll take a look at schedutils. The
> xfce-cpuload-plugin seems to be nice, but I'm kde-user :).

KDE user here too.  =8^)  I use ksysguard in the panel, with 4 CPU 
monitors, one for each of my dual cores on each of my Opteron 290s.  It's 
quite useful, as user/system/nice is broken out on each one, and I have 
portage set to nice 19, so I can monitor how much it's using vs. other 
things (a lot of X updating the screen with konsole, particularly with 
composite turned on, somewhat less with quadkonsole, which seems to be a 
bit more CPU efficient but with fewer bells and whistles, particularly 
per session when one is running more than one).

The dual 290s have certainly met and passed my expectations, with things 
running much smoother when multitasking than the dual 242s I was running 
before.  With 8 gig of memory to go with it and quad drives in md/
RAID-0/1/6, I've no reason at all to complain about performance or want 
more ATM.  I'm a happy camper! =8^)

The next upgrade is my eyes, Lasik, then back to the computer for dual 
flat-panels and a GPU upgrade as the prices on high resolution hopefully 
continue to fall.

-- 
Duncan - List replies preferred.   No HTML msgs.
"Every nonfree program has a lord, a master --
and if you use the program, he is your master."  Richard Stallman

-- 
gentoo-amd64@gentoo.org mailing list

Re: [gentoo-amd64] compiling on dual-cores - only 50% usage per cpu

[Peter Humphrey]

On Saturday 10 Nov 2007, Bernhard Auzinger wrote:

> it's not a important question, but has anybody of you noticed when
> compiling certain packages the load on both cpu's is only about 50%. I
> have the feeling that it happens with packages that can not be split up
> into two jobs. But in this case one cpu should be on 100% load and the
> other one at ~0%. Maybe someone of you has a good explanation for this
> behaviour.

I posted a question like this a few months ago, and I did a fair amount of 
detective work. Pretty much inconclusive, I'm afraid, but it seemed to be 
due to a motherboard chipset problem. This is a SuperMicro H8DCE; the 
problem only appeared when this replaced a faulty MSI board. I think the 
conversation ran in May, so you should be able to find it in the archive if 
you're interested.

(I'd noticed that the BOINC scheduler was trying to get two processes run at 
full load, one on each CPU, but according to /top/ and /gkrellm/ each was 
getting just 50%, sometimes one on each CPU and sometimes both on the same 
one. And /top/ seemed confused about which CPU was running which tasks!)

I don't run BOINC any more, for other reasons, so I haven't run into the 
problem recently. Kernel compilations seem to go OK, so it's all a bit of a 
mystery. I concluded at the time: "I'm left with a vague feeling of 
dissatisfaction from not knowing what's going on, and a suspicion that 
something in the nForce2 chipset doesn't match what the kernel thinks. 
Kernel versions seem not to affect this problem, at least not in the range 
of versions I could find."

-- 
Rgds
Peter.
Linux Counter 5290, Aug 93
-- 
gentoo-amd64@gentoo.org mailing list

Re: [gentoo-amd64] Re: compiling on dual-cores - only 50% usage per cpu

[Bernhard Auzinger]

> KDE user here too.  =8^)  I use ksysguard in the panel, with 4 CPU
> monitors, one for each of my dual cores on each of my Opteron 290s.

That sounds like you are having a infernal machine. 8 cores, that's not bad. 
What are the requirements for the power supply when running 4 Opteron 290s?

Rgds
Bernhard
-- 
gentoo-amd64@gentoo.org mailing list

Re: [gentoo-amd64] Re: compiling on dual-cores - only 50% usage per cpu

[Bernhard Auzinger]

> KDE user here too.  =8^)  I use ksysguard in the panel, with 4 CPU
> monitors, one for each of my dual cores on each of my Opteron 290s.

Oh, it's 4 cores, not 8. But not bad either.

Rgds
Bernhard
-- 
gentoo-amd64@gentoo.org mailing list

Re: [gentoo-amd64] compiling on dual-cores - only 50% usage per cpu

[Bernhard Auzinger]

Am Sonntag 11 November 2007 schrieb Peter Humphrey:
> On Saturday 10 Nov 2007, Bernhard Auzinger wrote:
> > it's not a important question, but has anybody of you noticed when
> > compiling certain packages the load on both cpu's is only about 50%. I
> > have the feeling that it happens with packages that can not be split up
> > into two jobs. But in this case one cpu should be on 100% load and the
> > other one at ~0%. Maybe someone of you has a good explanation for this
> > behaviour.
>
> I posted a question like this a few months ago, and I did a fair amount of
> detective work. Pretty much inconclusive, I'm afraid, but it seemed to be
> due to a motherboard chipset problem. This is a SuperMicro H8DCE; the
> problem only appeared when this replaced a faulty MSI board. I think the
> conversation ran in May, so you should be able to find it in the archive if
> you're interested.
>
> (I'd noticed that the BOINC scheduler was trying to get two processes run
> at full load, one on each CPU, but according to /top/ and /gkrellm/ each
> was getting just 50%, sometimes one on each CPU and sometimes both on the
> same one. And /top/ seemed confused about which CPU was running which
> tasks!)
>
> I don't run BOINC any more, for other reasons, so I haven't run into the
> problem recently. Kernel compilations seem to go OK, so it's all a bit of a
> mystery. I concluded at the time: "I'm left with a vague feeling of
> dissatisfaction from not knowing what's going on, and a suspicion that
> something in the nForce2 chipset doesn't match what the kernel thinks.
> Kernel versions seem not to affect this problem, at least not in the range
> of versions I could find."
>
> --
> Rgds
> Peter.
> Linux Counter 5290, Aug 93

I don't think that it is a motherboard issue because I watched the same 
behaviour not only on my amd64X2 but also at work at a core2-duo and older 
pentium4's.

I will try to find the thread from may.

Thank you.

Bernhard
-- 
gentoo-amd64@gentoo.org mailing list

[gentoo-amd64] Re: compiling on dual-cores - only 50% usage per cpu

[Duncan]

Bernhard Auzinger <e0026053@student.tuwien.ac.at> posted
200711111148.44058.e0026053@student.tuwien.ac.at, excerpted below, on 
Sun, 11 Nov 2007 11:48:43 +0100:

> That sounds like you are having a infernal machine. 8 cores, that's not
> bad. What are the requirements for the power supply when running 4
> Opteron 290s?

Only 4 cores. =8^(  The "2xx" indicates a max 2-way machine.  Previously, 
it was 2 x 242s, single 1.6 GHz cores.  Now it's 290s, dual 2.8 GHz 
cores, so 2 cores x 2 sockets, 4 cores total over 2 slots.  (Each core 
shows up as a CPU on the monitors, thus the 4 CPU monitors running above, 
actually 4 cores over two CPUs.)

I'd have liked to go with the 8xx series, but they run way more money per 
socket and the boards are of course higher as well.  Way out of my budget 
range, even for "incremental" upgrade.  Well, out of my range for a 
couple years anyway... it'll be mid-range in a couple years.

-- 
Duncan - List replies preferred.   No HTML msgs.
"Every nonfree program has a lord, a master --
and if you use the program, he is your master."  Richard Stallman

-- 
gentoo-amd64@gentoo.org mailing list