This is a multi-part message in MIME format.
--
[ Picked text/plain from multipart/alternative ]
Yeah, that is the problem. Since an HT system reports twice as many
processors, vmstat mis-reports maximum CPU usage. It sees four
processors reported, but in reality there are only two execution units.
It looks at cpu utilization across all four processors, adds it up and
divides by four. But since there are only two execution units, if all
four are in use, on average the maximum any one will get is 50%,
* Processor A execution unit ->100% utilization
o HT1 -> 50% utilization
o HT2 -> 50% utilization
* Processor B execution unit ->100% utilization
o HT1 -> 50% utilization
o HT2 -> 50% utilization
Add A(HT1-HT2), B(HT1, HT2) up and divide by 4 to get total CPU
utilization and you get 50%, when in reality both processors are at
100%. To really understand what the system load is, you need to run SAR,
get the by-CPU report out and add up the numbers yourself, recognizing
that you have to add the utilization of the paired HT units together to
calculate load and idle utilization for that processor, then add those
results together and divide by two to get the total system utilization.
Sid
e-Plutonia wrote:
>HT and the CPU usage change in % as I have observed running on Dual
>Xeon 3.2's seems to change by 4-5% avg / total on the system, but not
>really per proccess. I have stress tested Dual Xeon 3.2's which ran at
>5.0 to 7.0 load with only 50% CPU usage.
>
>On 8/18/05, Sid Stuart <[EMAIL PROTECTED]> wrote:
>
>
>>This is a multi-part message in MIME format.
>>--
>>[ Picked text/plain from multipart/alternative ]
>>One problem these days is that the statistics gathering/reporting tools
>>do not handle Intel's HT feature well. There was a thread a month or so
>>back where people were turning it off to get better statistics. I
>>wouldn't turn it off myself, it improves performance, but it does make
>>understanding the system load a bit harder.
>>
>>Sid
>>
>>m0gely wrote:
>>
>>
>>
>>>Michael McKoy wrote:
>>>
>>>
>>>
>>>>Maybe someone can confirm or deny, but I'm not certain I would trust
>>>>Windows's measurements.
>>>>
>>>>
>>>I wouldn't trust some Linux numbers either. Every now and then someone
>>>pops up on the list claiming a full 32 player server that runs a 5% cpu
>>>load.
>>>
>>>--
>>>- m0gely
>>>http://quake2.telestream.com/
>>>Q2 | Q3A | Counter-strike
>>>
>>>_______________________________________________
>>>To unsubscribe, edit your list preferences, or view the list archives,
>>>please visit:
>>>http://list.valvesoftware.com/mailman/listinfo/hlds_linux
>>>
>>>
>>>
>>>
>>>
>>--
>>/Sid Stuart
>>Games Engineering /
>>
>>Yahoo! Games
>> 701 First Avenue
>> Sunnyvale, CA 94089
>> (408) 349-7575 office
>>
>>
>>--
>>
>>
>>_______________________________________________
>>To unsubscribe, edit your list preferences, or view the list archives, please
>>visit:
>>http://list.valvesoftware.com/mailman/listinfo/hlds_linux
>>
>>
>>
>
>
>--
>DJ Fadyeyev
>Founder
>e-Plutonia
>
>_______________________________________________
>To unsubscribe, edit your list preferences, or view the list archives, please
>visit:
>http://list.valvesoftware.com/mailman/listinfo/hlds_linux
>
>
>
>
>
--
_______________________________________________
To unsubscribe, edit your list preferences, or view the list archives, please
visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux
