I had many helpful suggestions and thank you all,
suggustions included lsof, ps, collect and top.
None of these really provide what I was hoping for,
so perhaps here's a challange for HP of some enterprising
perl scripter.
What I want is a command that will tel me who and what
is creating I/O activity on a given file system, perhaps ...
iomonitor <AdvFSDomainName>
PID OWNER UID % of current I/O activity
=============================================
1 100 80
2 200 10
3 100 5
4 100 5
Never mind, dream on Lawrie :-)
Here are some words on the subject from Alan Rollow.
Not easily. Each process has an data structure associated
with it that collect resource usage data, but I'm not sure
this is available of ps(1). You might want to check all
the ps(1) options just to be sure. And track down a copy
of "top" to see if it can collect anything.
One thing you might try, is to put together a list of
suspect processes and send them a SIGSTOP signal. That
will stop most processes without killing them. If the
load goes down, you found a good suspect. Then send
the process a SIGCONT to let it run again.
If there are users associated with some of the suspect
processes you should check with them to see if they
reason to believe their process might be causing a
problem.
Some knowledge of the work-load change would also be
useful, since it offer a clue what changed. The obvious
change was either an increase in bandwidth usage or
request rate. The other interesting bit of information
is average I/O size. Whether it went up or down offers
some clue what might have changed.
Lawrie
Received on Wed Nov 12 2003 - 19:39:47 NZDT