I just found this old screenshot from one of my previous jobs. It was taken on December 9th 2003, while one of the web hosting servers went woowoo due to badly optimized web site. Load average went sky high to 682! Anyone else had such a high load before or am I the absolute champion? :)
PS: I was using Fluxbox back than! Wow, crazy youth! :)
Oh, I have seen one FreeBSD server surviving load over 800. “680″ is the second highest load I’ve ever heard of.
i got around 900 few times this year on few boxes :|
debugging it sux
I have seen a Solaris box hit both 700 and 1200 (roughly) but these numbers are not necessarily directly comparable to yours. Different operating systems report that number differently depending on how many processors you have.
I think those boxes would have both had 8 (4 dual core) but we also had some that had 24 and my memory is not accurate enough to be sure which one I was on.
Unfortunately, I didn’t have the presence of mind to take a screen shot.
I’ve had 2941. This was achieved by compiling a linux kernel with ‘make -j’ on a fairly massive ramdisk.
That server can sure use some mysql tuning…
http://imgur.com/f8WMM.png
Hehehe… seems this happens quite a lot. Nice to see it is not limited to Linux only. :)
@br41n, I was lucky enough to know what the problem was so I could react very fast. :)
@dave, this happened on a one CPU, PIII machine so there is nothing to divide here. :)
@niall, you really know how to kill a machine. Good job. :)
@julien, MySQL was also the problem in my case. Or rather bad usage of MySQL. One customer with a very popular web site stored images inside of MySQL – not references to images, but full freakin’ images! And since it was a web site for a gossip newspapers you can imaging how much images there were on each page! :)
Meni je zapravo fascinantan tvoj wallpaper :) Zaprepascujuce, kako se neke stvari uopste ne menjaju :P
I have had (but no screenshots) a system with load average of about 10000. It was a massive fork bomb caused by udev behavior when detecting ~4200 storage LUNs. If I am not mistaken, each udev detection process (one per each block device) has two children, so you can do the math here. The system had 16 physical cores (HT was disabled), and it was very sluggish during that time.
When I was able to login to it (it was part of a normal reboot process, the slow wake-up caused by the huge amount of storage LUNs), I have had the pleasure of capturing the ‘top’ command showing all processes on screen with the value ’9999′ on the CPU column. I was sure the cdrom will burst and I will be able to fill my hands with coins.
I did not :-/
Ez
4200 LUNs? Was someone playing with the zoning or what?