CPU - HP UX

This is a discussion on CPU - HP UX ; Hi I am a newbie to Unix and excuse me if this question had already been answered many times before. We have HP-Unix 11i running Oracle Database 9i. Here are some stats from UNIX box procs memory page faults cpu ...

+ Reply to Thread
Results 1 to 7 of 7

Thread: CPU

  1. CPU

    Hi
    I am a newbie to Unix and excuse me if this question had already been
    answered many times before.

    We have HP-Unix 11i running Oracle Database 9i.

    Here are some stats from UNIX box

    procs memory page faults
    cpu
    r b w avm free re at pi po fr de
    sr in sy cs us sy id
    11 1 0 2438925 2261128 93 71 0 0 0 0
    0 1339 23054 1473 21 7 72
    11 1 0 2438925 2261128 8 1 0 0 0 0
    0 3238 64756 5163 77 23 0
    11 1 0 2438925 2261073 7 0 0 0 0 0
    0 3100 66464 4998 82 18 0
    11 1 0 2438925 2261073 5 0 0 0 0 0
    0 3129 67129 4875 79 21 0
    11 1 0 2438925 2261073 4 0 0 0 0 0
    0 3237 66589 4966 80 20 0
    11 1 0 2438925 2261074 21 2 0 0 0 0
    0 3177 70690 4962 82 18 0

    Top shows IDLE ranges between 0% to 20% typically in a 5 minutes
    monitor. I was told we have enough RAM (no memory issue)

    >From the report above, is it possible to say the machines suffers from

    CPU bottleneck as the run procs (11) is way ahead of number of
    CPUs(which is 4)..

    and while monitoring against a shell script (which copied about 30G of
    data) I noticed the run procs went up to 17 at one point..

    Just wanted to get your thoughts , is it indication that we need to add
    CPUs or does that need furthur analysis

    Thanks


  2. Re: CPU

    As a rule of thumb, we normally know a problem exists when our runq
    exceeds '4'.

    We tend to be fairly conservative due to the fact that we're a very
    on-line, need fast access and response times company.

    We have seen stuff like this happen to us in one of our environments,
    but it normally ends up being an Oracle tuning issue. For instance
    Oracle is spending too much time trying to figure out what index to
    choose, etc. (you know what they say, ask Oracle the right question and
    it will ponder the meaning of life for you).


  3. Re: CPU

    Thanks belvilad for your reply.

    Actually you are correct that 90% of the time it boils down to ORACLE
    tuning..
    but here is the larger picture

    we run peoplesoft (only a limited tuning is possible)
    Same database files copied over to some other unix box (to development
    where there are atleast 5+ instances running) - runs faster (same
    plan..same query..same oracle autotrace statistics )

    just wanted to make sure we are not CPU bound in production box alone
    and hence was wondering about runprocs value in vmstat (in dev, its
    always less than 3)

    Another question is, if vmstat shows a higher run procs count (than CPU
    Count) but no waits (third column in vmstat) or blocked (2nd column) ,
    does that mean we are still okay?

    The OS admin said there are no memory issues (no paging,swapping ).
    thats another reason why I am looking at CPU processing power (whether
    it needs to beefed up).

    if anyone can post what steps should I take to test CPU load (and
    waits) that would be helpful

    Thanks again


  4. Re: CPU

    Let's see...

    Again, we too have seen differing results from Oracle between
    development and production, but often this boils down to the data in
    the db and the different queries (and volume quite frankly) between the
    environments.

    Personally, I'd be looking at data that is a bit more detailed here
    before jumping to any conclusions. For example, sar -Mu will tell you
    some of the same information your seeing with Vmstat, but will break it
    out based on CPU and tell you how much waiting on io you're doing (your
    cpu loads could be caused by something else like a disk bottleneck of
    filesystem misconfiguration).

    Ultimately high queue lengths mean that your system has quite a bit of
    processing backlock waiting on the CPU's and ultimitely might manafest
    itself into digruntled users.

    The cause of that backlog is what your trying to determine. Is it a
    misconfigured oracle environment? Does the CPU load go down at night
    when the users log off (maybe it's not oracle, but a Unix system
    problem). Is your NIC card overloaded on the production box, etc.

    Applications like top, Glance, perfux, etc might help you to dive a bit
    deaper into what exactly is causing the backlog.


  5. Re: CPU

    Thanks again for your time and help.

    "Again, we too have seen differing results from Oracle between
    development and production, but often this boils down to the data in
    the db and the different queries (and volume quite frankly) between the

    environments."
    - well good thought..but what I meant was, do a HOT backup of
    production database (file after file) and copy it over to development
    instance and bring it up there (data volume is ditto same ; the only
    difference is user-load..since I dont have "loadrunner" or similar
    testing tools, I cannot add more stress..but the development server
    anyway is overloaded with more than 10 other running instances..)

    in spite of all this, development runs faster than production (sort of
    funny why we call this as prod box :-) )


    "The cause of that backlog is what your trying to determine. Is it a
    misconfigured oracle environment? Does the CPU load go down at night
    when the users log off (maybe it's not oracle, but a Unix system
    problem). Is your NIC card overloaded on the production box, etc.


    Applications like top, Glance, perfux, etc might help you to dive a bit

    deaper into what exactly is causing the backlog.
    "
    - thats new info to me..thanks
    I will have to read and understand what is NIC and how to monitor
    I see top with %IDLE jumping every two seconds somewhere between 10% -
    60%
    Glance and perfux - dont know much about it..will have to research more

    Thanks again for your inputs


  6. Re: CPU

    NIC stands for Network Interface Card

    Glance and Perf/UX (you really just need measureware unless you're
    really interested in the GUI) are products available from HP. Glance
    is really for short term, right now, performance analysis, perf/UX
    (actually that's the old name and I can't remember off the top of my
    head the new one) is more for long term stuff.

    If it were me I'd be looking at disk performance (sar -d), network
    activity (you can use nettl, glance, and a few opensourced programs to
    do this), etc before jumping to any conclusions.

    top is an excellent tool to help determine the cause of your
    performance woes. Good Luck!


  7. Re: CPU

    belvilad wrote:
    > If it were me I'd be looking at disk performance (sar -d), network
    > activity (you can use nettl, glance, and a few opensourced programs
    > to do this), etc before jumping to any conclusions.


    For network performance statistics other than those reported by Glance
    and Measureware, I'd probably start with netstat and lanadmin before I
    went with nettl. Nettl would be more for packet tracing, and frankly,
    tcpdump may be a better tool for that.

    rick jones
    --
    denial, anger, bargaining, depression, acceptance, rebirth...
    where do you want to be today?
    these opinions are mine, all mine; HP might not want them anyway...
    feel free to post, OR email to rick.jones2 in hp.com but NOT BOTH...

+ Reply to Thread