Hi Ron:
Sorry for the late response. I had back to back conferences and was away
for a good chunk of the time, that's why I didn't get a chance to reply
until now. Also, in the future it would be great if you can email the
mailing-lists as well so that other users/developers could jump into the
discussion.
I have forwarded the email to ganglia-developers mailing-list, if you are
not subscribed, please do so.
dev Zero has already created a bugzilla entry, so I suggest you create a
unified diff against our code in trunk and upload it there so we can better
keep track of it:
http://bugzilla.ganglia.info/cgi-bin/bugzilla/show_bug.cgi?id=285
I am not really in a position to review the patches for AIX and was hoping
that Michael Perzl would be able to do so, however I am having a hard time
getting hold of him (perhaps my emails went to his spam folder?).
It would be great if the Ganglia AIX community could help review and test
the patches on different AIX platforms.
Thanks a lot for your contribution and interest for the Ganglia project.
Cheers,
Bernard
On Wed, Nov 17, 2010 at 6:48 AM, Ron Wellnitz <ron.welln...@debeka.de>wrote:
> Hi Bernand,
>
> i've found the time, to fix the calculation for the cpu usage in the
> metric.c module.
>
> The sourcecode was compiled on a AIX Power5 maschine with AIX 5.3 TL 11
> (5300-11-04-1015).
>
> I our environment i could test the new module on a (OS: AIX 5.3 TL 11):
>
>
> Power5 LPAR:
> System configuration: type=Shared mode=Uncapped smt=On
> System configuration: type=Shared mode=Uncapped smt=Off
> System configuration: type=Shared mode=Capped smt=On
> System configuration: type=Shared mode=Capped smt=Off
> System configuration: type=Dedicated mode=Capped smt=On
> System configuration: type=Dedicated mode=Capped smt=Off
>
> Power4 LPAR*:
> System configuration: type=Dedicated mode=Capped smt=Off
>
> *SMT and Shared CPU ar not supported on Power4 Hardware
>
>
> The values for cpu usage were allways correct.
>
> I've not yet included the Power5 and above, specific values like 'CPU pool
> usage' or 'Entitlement' statistics (the known extensions from Michael
> Perzl).
>
> We need too test the changes on other platforms and configuration (Power6,
> AIX 6.1, etc.).
>
>
> Regards,
> Ron
>
>
> --------------------------------------------------------------------------------------------------------------------------------------------------------
>
> Hi Bernard,
>
> i have started to spend some time into the issue but it is hard to fix,
> because its the first time for me working with c code.
>
> Currently i have a lot to do and no more time for rewriting the metrics.c.
>
> In the attachement, i have put the original IBM Code for calculating
> performance values (vmstat example code).
>
> By reading this example, the problem become clear...
>
> The current metrics.c calculate the performance values for the logical cpu
> usage but not for the physical one.
> The next problem is to determine, if the lpar use "entitled capacity" (cpu
> units from a shared processor pool).
> If so, the formula is different and the fix from dev Zero (replace
> 'perfstat_cpu_total' with 'perfstat_partition_total') will not help.
>
> We need to include the calculation formula from the given IBM examples,
> into "our" metric.c.
>
>
> Am 21.10.2010 09:49, schrieb Bernard Li:
>
> Hi all:
>
> I've been trying to get a hold of Michael to get his patches upstream but
> haven't heard back from him yet. It would be great if the AIX community
> could help me out by pointing out/providing patches that should be checked
> into our repository so we can fully support Ganglia on AIX platform.
>
> Thanks,
>
> Bernard
>
> On Wed, Oct 20, 2010 at 11:52 PM, Ron Wellnitz <ron.welln...@debeka.de>wrote:
>
>> Hi dev zero :)
>>
>> we use ganglia to monitoring our AIX boxes (AIX 5.3 TL 11) and saw the
>> same differences in the cpu usage/idle in the output from vmstat and gangia.
>>
>>
>> First you can try to use the AIX RPM packages from this site [
>> http://www.perzl.org/ganglia/ ]. I have not yet testet the latest version
>> (3.1.2), maybe it will fix the problem.
>>
>> In my case i replaced the default metrics from gmond by values from
>> vmstat.
>>
>>
>> Regards,
>> Ron
>>
>>
>> Am 20.10.2010 18:15, schrieb dev Zero:
>>
>> Have noticed that the metrics.c file has been coded using the
>> perfstat_cpu_total system call, and was wondering about the reasoning in
>> using the perfstat_cpu_total CPU stats in preference to those produced via
>> the perfstat_partition_total system call ?
>>
>> The reason I ask is that there seems to be a wide level of varience
>> between the CPU User and CPU Idle metrics produced by both system calls.
>> If someone understands the difference between the CPU metrics provided by
>> these two system calls, and why they would provide considerably different
>> CPU User and Idle metrics on a 2 CPU Aix 5.3 LPAR with no sharing of CPU's
>> involved, I would very interested to hear from you.
>>
>> The standard Aix performance tools (vmstat, sar, topas, nmon) all seem to
>> be using the metrics provided by the perfstat_partition_total. This means
>> (in my case) that Ganglia's CPU User and CPU Idle graphs do no match what is
>> shown to be happening on that LPAR with the standard Aix Performance tools.
>> (ie. Ganglia graphs show LPARs with 30% ish idle time, but vmstat etc
>> shows no idle time).
>>
>> Documentation on the perfstat_cpu_total system call seems to indicate that
>> the perfstat_partition_total system call should be used instead.
>>
>>
>> http://publib.boulder.ibm.com/infocenter/aix/v6r1/index.jsp?topic=/com.ibm.aix.basetechref/doc/basetrf1/perfstat_cputot.htm
>> Quote : All data consumers requiring processor use must be modified to
>> support PURR-based computations as shown in the example for the
>> perfstat_partition_total interface under Perfstat API programming.
>>
>> While I'm confident that there is a good reason for the use of
>> perfstat_cpu_total in the aix/metrics.c source, I would like to understand
>> the reasoning and so banish any nervousness in trusting the stats provided
>> the Ganglia product.
>>
>> Appreciate any information shedding light on this.
>>
>> Thanks and Best Regards to all.
>>
>> PS. My hat is off to the Ganglia Team/Community. :-) Very impressive
>> product.
>>
>>
>>
>> ------------------------------------------------------------------------------
>> Nokia and AT&T present the 2010 Calling All Innovators-North America contest
>> Create new apps & games for the Nokia N8 for consumers in U.S. and Canada
>> $10 million total in prizes - $4M cash, 500 devices, nearly $6M in marketing
>> Develop with Nokia Qt SDK, Web Runtime, or Java and Publish to Ovi Store
>> http://p.sf.net/sfu/nokia-dev2dev
>>
>>
>> _______________________________________________
>> Ganglia-general mailing
>> listganglia-gene...@lists.sourceforge.nethttps://lists.sourceforge.net/lists/listinfo/ganglia-general
>>
>>
>>
>> ------------------------------------------------------------------------------
>> Nokia and AT&T present the 2010 Calling All Innovators-North America
>> contest
>> Create new apps & games for the Nokia N8 for consumers in U.S. and Canada
>> $10 million total in prizes - $4M cash, 500 devices, nearly $6M in
>> marketing
>> Develop with Nokia Qt SDK, Web Runtime, or Java and Publish to Ovi Store
>> http://p.sf.net/sfu/nokia-dev2dev
>> _______________________________________________
>> Ganglia-general mailing list
>> ganglia-gene...@lists.sourceforge.net
>> https://lists.sourceforge.net/lists/listinfo/ganglia-general
>>
>>
>
------------------------------------------------------------------------------
Beautiful is writing same markup. Internet Explorer 9 supports
standards for HTML5, CSS3, SVG 1.1, ECMAScript5, and DOM L2 & L3.
Spend less time writing and rewriting code and more time creating great
experiences on the web. Be a part of the beta today
http://p.sf.net/sfu/msIE9-sfdev2dev
_______________________________________________
Ganglia-developers mailing list
Ganglia-developers@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/ganglia-developers
