[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: [OFFTOPIC] cpu idle time



> I was recently installing linux on a poor 386sx, with 4M of memory.
> 
> I noticed that during a compile of a kernel, 2.0.30, which took 35 hours
> to complete, vmstat showed a large idle time. I was swapping like mad, and
> came to the conclusion that time spent in swap is accounted for in the
> idle time?
> 
> AIX 3.2.5 has an additional field for the cpu percentages, called wait(ing
> for io)

Well, if you do the following before you start the kernel compile:

$ cat > c.c
main(){while(1);}
$ gcc -o c c.c
$ nice -n 20 c

Then I think you'll see the "Idle" time drop significantly during the
kernel compile. So, the trick is, the time while Linux is waiting
for the HD's to spin round, and return data, the kernel really wanted
to do something else (for example, execute the tight loop above).

Now, when you want to implement this "wait for io" field, I think the
output would be rather strange: If you don't run that while loop above,
then in the kernel compile the "Wait for io" field would be really high,
but if some user was running the while loop simultaniously (or some
computation), suddenly the kernel kompile would take equally long,
but now the it would seem asthough there was no "wait for io" time used.

I *think* that is why Linux doesn't have this "Wait for IO" field,
though I'm basically just guessing.

Being a Linux enthousiast, I also like to think that AIX has that
additional field, because AIX actually does "wait for IO", i.e., while
waiting for IO it doesn't do run any other jobs in the background.
If that is the case, then the AIX "Wait for IO" field does return reliable
information: information that tells how bad AIX is. But Actually,
I don't think AIX would be that bad. (No experience)



-- 
joost witteveen, joostje@debian.org

My spamfilter is so good, it correctly catches 90% of incoming spam,
*including* all email from my PhD supervisor.


--
TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to
debian-user-request@lists.debian.org . 
Trouble?  e-mail to templin@bucknell.edu .


Reply to: