Results 1 to 6 of 6
  1. #1
    Join Date
    May 2003
    Location
    France
    Posts
    112

    Question Unanswered: Strange system stats ...

    I have collected system stats with dbms_stats.gather_system_stats and display it :
    PNAME := 'cpuspeed';
    DBMS_STATS.GET_SYSTEM_STATS(status, dstart, dstop, pname, pvalue, stattab => 'mystats', statid => 'CTRL_REJET', statown =>
    'FRED');
    DBMS_OUTPUT.PUT_LINE('status : '||status);
    DBMS_OUTPUT.PUT_LINE('cpu in mhz : '||pvalue);

    status : COMPLETED
    cpu in mhz : 614

    The problem is that our system is a four 1200 Mhz cpu Sun (checked with psrinfo -v)

    Why Oracle don't collect the right cpu speed ?
    What could be the impact on performance ?

    Thanks for help

  2. #2
    Join Date
    Mar 2002
    Location
    Reading, UK
    Posts
    1,137
    Not sure, but I think Oracle CPU cycles are probably some normalized unit of work rather than actual CPU cycles. After all different CPUs will do different amounts of work for the same number of CPU cycles, so actual physical CPU cycles isnt particularly useful.

    Alan

  3. #3
    Join Date
    Aug 2004
    Location
    France
    Posts
    754
    Hello,

    This is not precisely the subject, sorry, but I have a question somewhat related to this thread :

    Did someone else notice (not with Oracle specifically : I didn't take time to test) that Sun servers are significantly slower than PCs ? We have a 600 MHz SunFire V120 with 2 Gb 133 MHz Memory, and some benchmarks (C programs) showed dramatic performance gains on a 3 GHz PC with 512 Mb RAM (MUCH LESS expensive). I plan to do some benchmarks with Oracle in the next weeks to see if it confirms, but I wanted to know if somebody already experienced such a phenomenon. Are Sun machines so slow ?

    Thanks and regards,

    RBARAER

  4. #4
    Join Date
    May 2003
    Location
    France
    Posts
    112
    thanks for responding AlanP

    The cpuspeed is in "cycles per second, in millions " so mhz (dixit Note:149560.1)

    ....

  5. #5
    Join Date
    Mar 2002
    Location
    Reading, UK
    Posts
    1,137
    With regards CPUspeed my twin 2.8 GHz P4s (with HT) show a CPUspeed of 794 so either it normalizes CPUspeed or it measures actual CPU cycles used during the time your gather system stats ran (also unlikely you will have 100% load on the CPU during this time).

    As regards Sun V120 dont forget this a bottom end sun CPU (UltraSparc IIi and it only runs at 650MHz). The latest Sun servers use UltraSparc IV at upto 1.2 GHz (with dual cores aswell) which would compete much better with a P4. Also Sun servers seem to scale much better as the number of CPUs increase IMO.

    Alan

  6. #6
    Join Date
    May 2003
    Location
    France
    Posts
    112
    Alan, for information, someone gives me these formula:
    This cycles are converted to the units of i/o cost using the following conversion formula:

    cost = CPU-RSC / (1000*CPUSPEED*SREADTIM)

    where:
    CPU-RSC is the estimated number of required cpu cycles (you can get them from the event 10053 trace - CBO trace)
    CPUSPEED - the speed of CPU
    SREADTIM - average single block read time

    So, you are right

    Thanks for your help

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •