Bizzare Performance monitor results

[jerry2]

Hi there

Sorry if this is not directly related to Plesk, but maybe it is as my old server had plesk 7.5 and new one has 8.1.... I am not sure any more..

I had a server Windows 2003 Web (low cost server, not high end) and the performance monitor was quite ok for my site (it is quite big)...

Now I moved to another Windows 2003 Web server with Opteron Dual 1212 CPU, same RAM (2 GB) and 2 x 200 GB HD Samsung Spinpoint 200 series...

Now site is working great, my CPU is down from 30-50% to 10-25% and pages/sec is not more than on the old site (averga about 3, but can peak to 38). I say this is OK, as I have a mysql database that is not perfectly optimised yet...

But what makes me worry is the Avg. Disk Queue Length monitor. The counter on graph seems quite high jumping from 10 to 100, but the numbers are: Average 0.4-0.6 but maximum is 429... I mean??!?!? 3 is supposed to be bad, but 429??? WHat the heck is going on here? But the Last is never so high as maximum says (I have a 1 sec freeze point)...

Now I did research and if pages/sec is high and Avg. Disk Queue is high I have a disk bottleneck. But it doesn't seem so. THe disk Queue means as far as I know the queue of tasks to be written (lag) and 429 is certanly a bad value... But average is 0.5, that from the documentation seems ok.

I don't know what to think of this readings... Could Dual CPU be the reason? Or that I have a software RAID 0 of the 2 disks? But I have broke the mirror and still high peaks although lower. I set some other parameters and seems WRITE is the problem not read, write jumps to 429 (so it says) and it happens on both disks the same...

Could this be a hardware malfunction or should I ignore altogether untill Avg. Disk Queue Length is above 2 in average?

Any healp would be appreaciated.

Yours

Jerry

 

[kpisman]

Hi.
Long disk queue length is not as bad as you think... relay RAID array will work as soon as queue lenght will become as long as heads number multiplied by something like 2, in better case.

Thus raid may be the case. To ensure RAID works system may make queue longer.

Other possible case is more RAM - as you have RAM left free OS may utilize it it to write cache. Windows always attempts to use lazy write.

What is dis performance is about for Plesk - is avg disk idle time. This shows as much throughput you have left.

Also you can try read SMART info from your drives, to ensure drives are ok.

 

[jerry2]

Thank you for the info. The number 429 must be a mistake IMHO. You can not have 429 at that pointer.

What I did yesterday is changed the 2 RAID driver , switched physically their places (ok only the cables)... and now max peak is 1.9 )) But they are same company drivers, so... wierd.

 

[kpisman]

Mostly - this was trouble of perfcounters themselves.
In certain situations even %Cpu use may be more than 100.
In this case - again look at relative disk queue length before and after your changes.

 

[jerry2]

avg. disk idle time is still same, about 90%. avg. disk queue still has peaks to 100 on scale, but now they read 1.8 and before they read 429,489 This was a perf. counter mistake. But nothing changed only I swapped both disks.