Re: Diagnosing performance problems

From: Jaime Cardoso (jaimec_at_solsuni.pt)
Date: 12/20/03


Date: Sat, 20 Dec 2003 03:11:36 +0000


<snip>

>
> But when I use the process monitoring utiltiies like top, prstat, ps -o
> vsz, etc, they show each process using 250M for a grand total of some
> ridiculous number like 100GB or something.
>
> Dont these utilities understand the concept of shared memory segments
> and adjust the numbers accordingly?

Nope, they don't

There is a "new" Sun Blueprint called Performance Forensics that
explains a little about this.

In your case, I would start in thinking if there was some suddent
performance degradation or, things were just getting worst from time to
time.

 From your vmstat, I get that the sr collumn is 0, this is good, it
shows your memory isn't tooooooooo bad.
You page in and page out's are to big, this may suggest you Oracle is
requesting lot's of info in bursts, see Oracle configuration (and your
sql to see what's happening). Or you may have a memory shorness (or a
memory leak with, in this case, reboot).
Has many people have said, your disks are too buzy. Stiping your
Tablespaces across several disks will improve your performance (why not
use c2t3d0, c4* and/or c5* ?)
Your access to the disks are mostly reads so, you can always create some
Oracle indexes and / or mirror your data. This will improve your
reliability and, make the reads in round robin for each side of the mirror.

PS. I just saw something that can't be right.
your iostat -xn is actually an iostat -xn 5 and you copied here the
second output thie command showed, right?

The first output is the accumulated statistics since the machine booted
and, are completely useless. Please confirm this because this may change
all that was said here. (the same is valid for top, vmstat, prstat, etc,
etc).



Relevant Pages

  • Re: Diagnosing performance problems
    ... > But when I use the process monitoring utiltiies like top, prstat, ps -o ... shows your memory isn't tooooooooo bad. ... your disks are too buzy. ...
    (comp.sys.sun.admin)
  • Re: Linux 2.6.29
    ... Ouch - 480 seconds, how much memory is in that machine, and how slow ... The 480 secondes is not the "wait time" but the time gone before ... that delay happened. ... The disks are a Nexsan SataBeast with 42 SATA drives in ...
    (Linux-Kernel)
  • Re: I/O performance troubleshooting
    ... Is there a fr:sr ratio I should be looking for? ... So in my case my memory is not over-committed and I should alocate more of it to Oracle. ... It seems pretty obvious to me my performance problem is with the disks I have. ...
    (AIX-L)
  • Re: I/O performance troubleshooting
    ... You are running out of memory a lot, but not paging as a result. ... The comments on the SGA look dead on to me. ... I should alocate more of it to Oracle. ... It seems pretty obvious to me my performance problem is with the disks I have. ...
    (AIX-L)
  • Re: alpha box and licence too? - Suggestions
    ... > The TC systems are old and relatively slow, ... The DEC 3000 machines also have METAL CASINGS. ... > these systems shipped with 32-64 MB of memory. ... > compatible storageworks disks are all fine provided you obtain the right ...
    (comp.os.vms)