May 10th, 2007

Speaker Rabbit

Полиномиальный анализ по irc

<tridge> its not 10k processes, just 3600
<vl> not the c10k problem, the p10k problem :-)
<tridge> only 2647 of them hung ....
<tridge> the rest managed to exit OK
<tridge> i suspect there is a O(n^2) bug
<tridge> it does seem to be very very slowly recovering
<tridge> its down to 2643 hung now
<tridge> 2640 ....
<tridge> thats a lot of cpu cycles per exit() call
<tridge> for a 4 cpu xeon box :)
<vl> Sounds like the old solaris fcntl bug :-)
<tridge> this is truly amazing syscall latency :)
<tridge> 2610
<tridge> its going faster now
<tridge> 2467 ....
<tridge> 1876 ...
<tridge> 1009 ...
<tridge> hmm, stuck at 1009 :(
<vl> You should be able to measure the speed and do a polynomial analysis over it...
<tridge> ahh, its moving again
<tridge> 999 ...
<vl> Is it really o(n^2), or even o(e^n) :-)
<tridge> dunno!
<vl> While waiting you have plenty of time :-)
<tridge> we can do it from the irc log ...
<vl> LOL :-)
<tridge> 701
<tridge> 290
<tridge> 1
<tridge> yay!