Historical cpu consumption information?

[Wendell Nichols]

Mike McCarty wrote:
> Wendell Nichols wrote:
>   
>> My fedora 10 laptop routinely "freezes".  That means the menu's and 
>> applications don't respond to mouse clicks.  After a few seconds (or 
>> sometimes many seconds) it frees up and works normally again.  If I look 
>> at the cpu consumption graph I can see that something had the processer 
>> pegged at 100% but what?  during the episode there is no way to switch 
>> to top or the ksysguard process list because the system is 
>> unresponsive.  So how do I find out what caused the lockup?
>> wcn
>>     
>
> Here's a possibility.
>
> $ while true; do date>>time.log; ps g -f --sort=utime,stime | head -n 3 
>  >> time.log; sleep 1; done
>
> As root, use renice to set the process to run at high priority;
> it might help. Perhaps you'll need to use nice in the loop to
> force things to run.
>
> That file will grow fairly rapidly, so you might want to set up another
> one to delete the file once an hour or day or so. You may want or need
> to tweak the number of lines you capture, and how often it runs.
>
> When an "event" occurs, you could kill the process, and
>
> $ tail -n 24 time.log
>
> to see the last 6 entries. To look at other entries, you could use
> something like
>
> $ tail -n 40 time.log | head -n 24
>
> if you've got a few at the end to omit. The example shows you
> what happened 10 seconds ago. This might save time loading
> it into a text editor or whatever.
>
> $ tail -n 80 time.log | less
>
> Might be good. Anyway, once you've got the data, you'll think
> of ways to proceed.
>
> Crude, but it may get you what you want.
>
> You got the suggestion to run top, but running top in batch mode
> results in a very rapidly growing log file, possibly too rapidly.
> I don't know of a way to limit the output top produces.
>
> Mike
>   
Thanks everyone for your input.  In the end I did a:
while [ 1 ]
do
date
ps -eo pcpu,pid,user,args | sort -k 1 -r | head -10
sleep 2
done
 
I ran that script with its output redirected to a log.  My machine then 
acted up (badly) and I was able to see that udevd was using 50% (one 
whole core) and a bunch of kde apps were unusually elevated ... in the 
12-20% area.  The machine ran like that for half an hour and finally 
thermaled off.  I had to cool it with a fan to get it restarted!
To make a long story short it seems to happen when I plug my blackberry 
in to charge!  I've seen some chat around the net and the fault is 
probably with the berry_charge module. 
I will find an electrical socket to plug that piece of junk into... (its 
only working function is to prematurely terminate conversations anyway).
Thanks again...
wcn