-----Original Message-----
From: James [mailto:[EMAIL PROTECTED]]
Sent: Tuesday, January 30, 2001 9:44 AM
To: [EMAIL PROTECTED]
Cc: [EMAIL PROTECTED]
Subject: Re: Killing "D" process
----- Original Message -----
From: "Matthew Melvin" <[EMAIL PROTECTED]>
To: "James" <[EMAIL PROTECTED]>
Cc: <[EMAIL PROTECTED]>
Sent: Tuesday, January 30, 2001 5:19 AM
Subject: Re: Killing "D" process
> Generally speaking a process in D state (disk) is 'in kernel' and theres
> nothing you can do about it. Sometimes you can attach strace to the
process
> and see what it's waiting on and give it what it wants (bring back it's
nfs
> mount for instance) but this is by no means reliable. lsof is also useful
> here, but sometimes you'll find it hangs for the same reason the process
> you're investigating is huhg.
>
There are 2 main types of load: actual cpu load, i.e. no idle time, and then
there is process load, where you have processes waiting on the cpu. (this
can be caused by many different things) They are not always related. We have
several hard hit web servers they get very high process loads but the cpu's
will be only utilized to 10%. (these are dual PIII 700 machines) Using the
output of 'uptime' you can grep and awk usefull information and log it.
#uptime
10:49am up 4 days, 1:05, 2 users, load average: 0.01, 0.06, 0.11
^^^^^^^^^^^^^^^^ (these
are 1, 5, and 15 minute readings)
Lastly, you can get high internal kernel contention which is very hard to
detect. This happens with servers that use a lot of threads and etc... BEA's
weblogic server is a prime example. When this happens no amount of
memory/cpu can improve this situation. You have reached the threshold for
that kernel/OS. (we load tested weblogic on Sun and Linux boxes for 6 solid
months, trying several different jdk's, etc.. and actually graphed the
different plateau's vs cpu vs memory, it was quite interesting) We also had
some of Sun's sr level engineer's working with us and used some of this
research for some white papers on 32bit vs 64bit mode under Solaris 7)
Also, I would gather load metrics much more often than 6hrs, I do it in 15
minute intervals.
Hope this helps,
CC
_______________________________________________
Redhat-list mailing list
[EMAIL PROTECTED]
https://listman.redhat.com/mailman/listinfo/redhat-list
_______________________________________________
Redhat-list mailing list
[EMAIL PROTECTED]
https://listman.redhat.com/mailman/listinfo/redhat-list
