UNIX for Advanced & Expert Users

i have a few mysql servers running, each one gets exactly the same amount of real world load (thousands of queries per second)

back when each one was running lenny, they all had very close load averages, decided to upgrade to squeeze, and all of a sudden load average is reading near 0.00 the entire time, but yet it is STILL processing the same amount of traffic it was before the upgrade, which means something is definitely not right

---------- Post updated at 02:01 AM ---------- Previous update was at 01:59 AM ----------

i tried ubuntu, same problem with a 2.6.37 kernel, then tried squeeze with upgraded 2.6.38 kernel, same problem again. if i downgrade squeeze to lenny kernel, load starts registering correctly again

That's it?

It's possible you're right, but we have no more idea what's stopped working than you do. Low load average doesn't tell us what's not happening -- how could it?

Test things. Make sure everything's responding. Put a specific load in and see if the right numbers come out. If you find anything wrong, then you have something you can get help with. If everything's working, rejoice, the kernel upgrade has improved your system!

Changes in the kernel can mean big improvements to scheduling. It may also have been configured with different time slice options and the like.

how can i compare between 2 systems the differences which may be affected?

---------- Post updated at 03:38 PM ---------- Previous update was at 02:15 AM ----------

OK, here is video of exactly what's going on...

wrong load averages - YouTube

Moderator's Comments:

I think you could have used a [youtube] BB tag ...

Depends what they're doing. Okay...

Would it really have been that hard to say "low load average during high cpu use"? Everyone wondered why you thought low load was a problem over and over but you never mentioned. Three words would have saved you the trouble of uploading a video, gotten you better responses days earlier, and not made everyone think you were insane.

How many cores does this computer have? One thing that might have happened is the old kernel was only using one core, and the new kernel sees all the cores -- the load average would be divided by the number of cores. cat /proc/cpuinfo in each kernel to see how many cores it thinks you have.

I wouldn't know that they changed the way the load average is calculated, but 2 things did catch my eye:
But again: you've given us a description of the differences and why you think this is happening, but you still haven't told us why this is a problem/bug? After all, it's probably only that they've adjusted the scheduling algorithm and thus reduced the load, meaning a snappier system. That's a good thing in my book.

for illustration purposes its 2 different servers, it's not hardware specific

it's not an optimization trust me, something is wrong, for some reason not many people notice it, couple of other incidents:

https://bugs.launchpad.net/ubuntu/+s...ux/+bug/513848

#620297 - base: vmstat and /proc/loadavg disagree - Debian Bug report logs

Then it's not a helpful comparison. "looks bad" isn't anything you can test.

The kernel patch you found might be a start.