In comp.unix.shell Olumide <50295@web.de>:
> Hi -

> I'd like to monitor the activity of the CPUs on a Linux machine. Is
> there a utility that does this? ... I know of mstat, but I'm looking
> for a utility that reports CPU activity dynamically -- like top.

What's wrong with 'top'?

--
Michael Heiming (X-PGP-Sig > GPG-Key ID: EDD27B94)
mail: echo zvpunry@urvzvat.qr | perl -pe 'y/a-z/n-za-m/'
#bofh excuse 69: knot in cables caused data stream to become
twisted and kinked

"Olumide" <50295@web.de> writes:

> I'd like to monitor the activity of the CPUs on a Linux machine.

gkrellm

--
Maurizio Loreti ROT13:ybergv@cq.vasa.vg | (@_
Un. of Padova, Dept. of Physics, Padova, Italy | //\
http://www.pd.infn.it/~loreti/mlo.html | V_/_

On Mon, 11 Dec 2006 07:32:15 -0800, Olumide wrote:

> Hi -
>
> I'd like to monitor the activity of the CPUs on a Linux machine. Is
> there a utility that does this? ... I know of mstat, but I'm looking
> for a utility that reports CPU activity dynamically -- like top.
>
> Thanks
>
> - Olumide

If you are looking for a graphical display then use gnome-system-monitor,
if you want the output in a terminal then the thing most like top is top.

Olumide wrote:
> Hi -
>
> I'd like to monitor the activity of the CPUs on a Linux machine. Is
> there a utility that does this? ... I know of mstat, but I'm looking
> for a utility that reports CPU activity dynamically -- like top.

It would help if you said what it is you need that "top" doesn't give
you.

Perhaps "sar"? See http://aplawrence.com/Words/2004_01_11.html

--
Tony Lawrence
Unix/Linux/Mac OS X Resources
http://aplawrence.com

> I'd like to monitor the activity of the CPUs on a Linux machine.

I'm interested in monitoring multiple CPUs

[ Followup-To -> comp.os.linux.hardware, well it doesn't match
that great but all other do even less... ]

In comp.unix.shell Olumide <50295@web.de>:
>> I'd like to monitor the activity of the CPUs on a Linux machine.

> I'm interested in monitoring multiple CPUs

Then finally start reading the fine manual 'man top' how to
switch on separate numbers for each CPU, trust me it is magic.

Good luck

--
Michael Heiming (X-PGP-Sig > GPG-Key ID: EDD27B94)
mail: echo zvpunry@urvzvat.qr | perl -pe 'y/a-z/n-za-m/'
#bofh excuse 391: We already sent around a notice about that.

Michael Heiming writes:

> In comp.unix.shell Olumide <50295@web.de>:
>> Hi -
>
>> I'd like to monitor the activity of the CPUs on a Linux machine. Is
>> there a utility that does this? ... I know of mstat, but I'm looking
>> for a utility that reports CPU activity dynamically -- like top.
>
> What's wrong with 'top'?

On (at least) Linux top(1) and i'd venture a guess every other utility
that uses `/proc/stat' or `/proc/uptime' is not very accurate, or goes
against my intuition of what constitutes load and measures that.

For instance: http://www.boblycat.org/~malc/apc/lo...-lbreakout.png
(substitute x86 for ppc to witness even more deviation on PowerPC)

The calculations of load based on timing the idle handler (and/or using
the niced background process and measuring it's forward progress) is in
stark contrast with what `/proc' tells.

(On the screenshot(s) red represents results obtained from `/proc' and
yellow is derived from measured time spent in idle handler)

The situation with load on Windows is similar, i have been so far
unable to find a method to measure the load the matches my, perhaps,
naive understanding of it. There is:
http://cpu.rightmark.org/products/rmclock.shtml
but there are no sources to see what it does, nor an explanation of
the algorithm(s) it uses.

--
vale

Olumide wrote:
>> I'd like to monitor the activity of the CPUs on a Linux machine.
>
> I'm interested in monitoring multiple CPUs
>

nmon for Linux is quite nice.

http://www-128.ibm.com/developerwork...u-analyze_aix/

-James

Olumide wrote:

> I'd like to monitor the activity of the CPUs on a Linux machine. Is
> there a utility that does this? ... I know of mstat, but I'm looking
> for a utility that reports CPU activity dynamically -- like top.
>
> Thanks

No offense, but this is the worst question I've seen on USENET in
months.

DS

"Olumide" <50295@web.de> writes:

> Hi -
>
> I'd like to monitor the activity of the CPUs on a Linux machine. Is
> there a utility that does this? ... I know of mstat, but I'm looking
> for a utility that reports CPU activity dynamically -- like top.

I like gkrellm - for a small footprint graph-type utility.


--
Sending unsolicited commercial e-mail to this account incurs a fee of
$500 per message, and acknowledges the legality of this contract.

In comp.unix.shell malc :
> Michael Heiming writes:

>> In comp.unix.shell Olumide <50295@web.de>:
>>> Hi -

>>> I'd like to monitor the activity of the CPUs on a Linux machine. Is
>>> there a utility that does this? ... I know of mstat, but I'm looking
>>> for a utility that reports CPU activity dynamically -- like top.

>> What's wrong with 'top'?

> On (at least) Linux top(1) and i'd venture a guess every other utility
> that uses `/proc/stat' or `/proc/uptime' is not very accurate, or goes
> against my intuition of what constitutes load and measures that.

> For instance: http://www.boblycat.org/~malc/apc/lo...-lbreakout.png
> (substitute x86 for ppc to witness even more deviation on PowerPC)

Works fine for me, perhaps you want to update procps to something
more recent, quite a few distro come with aged versions which
might not work that great on recent kernel.

--
Michael Heiming (X-PGP-Sig > GPG-Key ID: EDD27B94)
mail: echo zvpunry@urvzvat.qr | perl -pe 'y/a-z/n-za-m/'
#bofh excuse 99: SIMM crosstalk.

Michael Heiming writes:

> In comp.unix.shell malc :
>> Michael Heiming writes:
>
>>> In comp.unix.shell Olumide <50295@web.de>:
>>>> Hi -
>
>>>> I'd like to monitor the activity of the CPUs on a Linux machine. Is
>>>> there a utility that does this? ... I know of mstat, but I'm looking
>>>> for a utility that reports CPU activity dynamically -- like top.
>
>>> What's wrong with 'top'?
>
>> On (at least) Linux top(1) and i'd venture a guess every other utility
>> that uses `/proc/stat' or `/proc/uptime' is not very accurate, or goes
>> against my intuition of what constitutes load and measures that.
>
>> For instance: http://www.boblycat.org/~malc/apc/lo...-lbreakout.png
>> (substitute x86 for ppc to witness even more deviation on PowerPC)
>
> Works fine for me, perhaps you want to update procps to something
> more recent, quite a few distro come with aged versions which
> might not work that great on recent kernel.

My point is: it's the values in `/proc' which are in my view are false
not the utilities that present some graphs, cpu load percentages.
[and fwiw PPC version used for making the second screenshot uses 2.6.18.3
kernel and very recent procps]

The reason why it may work fine for you is it you are measuring
processes wich load the system in a sampleble (and prefferable by the
kernel) pattern.

You can try it yourself get LBreakout 010315 start a new game and
don't do anything, `top' `vmstat' will (most likely) tell you that
your system is ~1-3% loaded, while reniced background task will make a
lot less progress than it should have been if that was true, and
actually timing the idle handler will tell similar story - machine is
MUCH more loaded.

--
vale

"Olumide" <50295@web.de> wrote:
>
>> I'd like to monitor the activity of the CPUs on a Linux machine.
>
>I'm interested in monitoring multiple CPUs

One thing to remember is that there is no way to get an instantaneous
snapshot of CPU use. At any given point in time, a CPU is either 100% in
use or 100% idle. The only way to measure CPU use is to integrate over
time, which is what "top" reports.
--
Tim Roberts, timr@probo.com
Providenza & Boekelheide, Inc.

malc wrote:

> The calculations of load based on timing the idle handler (and/or using
> the niced background process and measuring it's forward progress) is in
> stark contrast with what `/proc' tells.

Maybe it is because for energy efficiency reasons, there is no mor
"background process". If a CPU is idle, it should stop.

You can't use a stopped CPU to measure anything.

--
Josef Möllers (Pinguinpfleger bei FSC)
If failure had no penalty success would not be a prize
-- T. Pratchett

On 2006-12-12, Josef Moellers wrote:
> malc wrote:
>
>> The calculations of load based on timing the idle handler (and/or using
>> the niced background process and measuring it's forward progress) is in
>> stark contrast with what `/proc' tells.
>
> Maybe it is because for energy efficiency reasons, there is no mor
> "background process". If a CPU is idle, it should stop.

How could you tell?

> You can't use a stopped CPU to measure anything.

If you're measuring something, it isn't stopped, it's doing your
measuring.

--
Chris F.A. Johnson, author
Shell Scripting Recipes: A Problem-Solution Approach (2005, Apress)
===== My code in this post, if any, assumes the POSIX locale
===== and is released under the GNU General Public Licence

Chris F.A. Johnson wrote:
> On 2006-12-12, Josef Moellers wrote:
>
>>malc wrote:
>>
>>
>>>The calculations of load based on timing the idle handler (and/or using
>>>the niced background process and measuring it's forward progress) is in
>>>stark contrast with what `/proc' tells.
>>
>>Maybe it is because for energy efficiency reasons, there is no mor
>>"background process". If a CPU is idle, it should stop.
>
>
> How could you tell?

By looking into the data sheets. By reading what is published. A stopped
CPU consumes less power than a running CPU.

>>You can't use a stopped CPU to measure anything.
>
>
> If you're measuring something, it isn't stopped, it's doing your
> measuring.

Exactly.

--
Josef Möllers (Pinguinpfleger bei FSC)
If failure had no penalty success would not be a prize
-- T. Pratchett

"Olumide" <50295@web.de> writes:

>> I'd like to monitor the activity of the CPUs on a Linux machine.
>
> I'm interested in monitoring multiple CPUs

I want to monitor my monitor...


Best regards
Martin Jørgensen

--
---------------------------------------------------------------------------
Home of Martin Jørgensen - http://www.martinjoergensen.dk

Josef Moellers writes:

> malc wrote:
>
>> The calculations of load based on timing the idle handler (and/or using
>> the niced background process and measuring it's forward progress) is in
>> stark contrast with what `/proc' tells.
>
> Maybe it is because for energy efficiency reasons, there is no mor
> "background process". If a CPU is idle, it should stop.

Linux calls default_idle/pm_idle[1] as part of it's normal idle chain,
default_idle/pm_idle might execute `hlt' or use something else to
reduce power consumption, in either case time spent doing this is
properly caclulated.

>
> You can't use a stopped CPU to measure anything.

I beg to differ:

start_time = gettimeofday ();
call reduce_power_consumption ();
end_time = gettimeofday ();
cummulative_idle_time += end_time - start_time;

is valid regardless of the way `reduce_power_consumption' works. And
this, in essence, is what i do.

And once again, the load reported by this is consistent with
measurements via background nice process that just burns cycles,
i.e. the machine (should) _never_ enters low power modes.

[1] On X86, things on other systems are a tad bit different.

--
vale