Andreas Baer wrote:
Bill Davidsen wrote:
One other oddment about this motherboard, Forgive if I have over-snipped this trying to make it relevant...
Andreas Baer wrote:
Willy Tarreau wrote:
On Mon, Jul 25, 2005 at 03:10:08PM +0200, Andreas Baer wrote:
There clearly is a problem on the system installed on this machine. You should
use strace to see what this machine does all the time, it is absolutely not
expected that the user/system ratios change so much between two nearly
identical systems. So there are system calls which eat all CPU. You may want
to try strace -Tttt on the running process during a few tens of seconds. I
guess you'll immediately find the culprit amongst the syscalls, and it might
give you a clue.
I hope you are talking about a hardware/kernel problem and not a software
problem, because I tried it also with LiveCD's and they showed the same results
on this machine.
I'm not a linux expert, that means I've never done anything like that before,
so it would be nice if you give me a hint what you see in this results. :)
Am I misreading this, or is your program doing a bunch of seeks not followed by an i/o operation? I would doubt that's important, but your vmstat showed a lot of system time, and I just wonder if llseek() is more expensive in Linux than Windows. Or if your code is such that these calls are not optimized away by gcc.
I don't know what exactly produces this _llseek calls, but I ran the compiled binaries on both machines (desktop + notebook) without any recompilation and so I think they should do the same (even if this is bad or not optimized), but I see a time difference of more than 2:30 :) This _llseek calls also don't seem to be faster or slower if you compare the times on the notebook and the desktop.
If the program and test data is not proprietary, would it help to have me run the test on my P4P800, P4-2.8, HT on, and see if that's an issue with your particular board or BIOS? I have the 1086 BIOS from my notes on that machine, I think you were running a later BIOS? 1091 or so, from memory?
Anyway, I would run a test that takes 3 minutes if it helps as a data point.