My God! I must have been the only person in the free world who did not
know the answer for this! Christopher Morris was right, I got absolutely
swamped with replies. I figure it must have been close to the whole list
membership. But, in defense of my pathetic existence, I found no mention
in the man pages of vmstat or iostat about this. You'd think that there
would be a reference to it at the bottom like all the rest. Oh well.
The answer was 'mpstat'. Thanks to about 57 people for that one. Several
other respondants noted that proctool would also monitor multiple cpu's. I
haven't tried that yet but I am getting it to give it a shot.
As a bonus (and a shallow attempt to make up for my ignorance), I will also
summarize the issue with Oracle. If you are running a client server setup
you can tune the Oracle application somewhat. Oracle is multi-threaded but
not in the standard definition. For Oracle, it means that a server process
is spawned for each connection rather than one server process handling all
the connections. It will take advantage of multiple processors but on a
process by process basis only, not by threads. Meaning a single process
cannot be spread across multiple processors. As I understand it, it's
basically asymetric multiprocessing. So, in this scenario, you have 2
options; tune the app so it sees all the cpu's and write the code so that
it uses parallel queries to spawn multiple processes. If you run a 3 tier
setup (like me) you have only one option, and that's in the code since the
app is in the second tier. There's nothing else to be done either by app
or OS or system.
Thanks to everyone, all 73 of you :-}
---------------------- Forwarded by Jeff Kennedy/NDS on 03/24/99 07:36 AM
"Jeff Kennedy" <email@example.com> on 03/23/99 10:53:01 AM
cc: (bcc: Jeff Kennedy/NDS)
Subject: Multi-cpu monitoring
Greetings and salutations,
I have an E3500 with 4 processors and I want to monitor each one
individually. I have been running vmstat and perfmeter to get the average
across all 4 but would like to see what each one is doing and how well the
load is being balanced between them.
Background (in case you care):
Oracle 7.something on Solaris 2.6. Running 5 different queries from remote
machines pegs the average cpu usage to about 80%. This was overnight when
no other processes were being run. We stopped the queries this morning to
restart the database in case of some wayward oracle processes. Once the
queries were stopped the cpu went back to 1%.
I just want to be able to see each cpu load as these queries come in and
how they balance.
This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:13:16 CDT