Re: Significant discrepancy in V$OSSTAT since upgrading to 10.2.0.5

From: Mladen Gogala <no_at_email.here.invalid>
Date: Wed, 21 Jul 2010 21:07:18 +0000 (UTC)
Message-ID: <pan.2010.07.21.21.07.17_at_email.here.invalid>

On Wed, 21 Jul 2010 12:40:21 -0700, Steve Howard wrote:

> On Jul 21, 11:51 am, Mladen Gogala <n..._at_email.here.invalid> wrote:

>> I used V$OSSTAT for monitoring in 10.2.0.4 and it was much more
>> accurate than is the case with 10.2.0.5. Look at this:
>>
>> SQL> column stat_name format a20
>> SQL> select stat_name,value from v$osstat
>>   2  where stat_name like '%TIME';
>>
>> STAT_NAME                 VALUE -------------------- ----------
>> IDLE_TIME              80163480 BUSY_TIME              42444869
>> USER_TIME              32543474 SYS_TIME                7838644
>> IOWAIT_TIME             8463912 NICE_TIME                     1
>> RSRC_MGR_CPU_WAIT_TI          0
>> ME
>>
>> SQL> select 42444869/80163480 from dual;
>>
>> 42444869/80163480
>> -----------------
>>        .529478872
>>
>> SQL> !sar -u 3 5
>> Linux 2.6.9-22.ELsmp (oracle13)         07/21/2010
>>
>> 11:45:27 AM       CPU     %user     %nice   %system   %iowait     %idle
>> 11:45:30 AM       all     17.83      0.00      8.08      6.00     68.08
>> 11:45:33 AM       all     19.23      0.00      8.83      2.66     69.28
>> 11:45:36 AM       all     14.51      0.00      6.67      0.83     77.98
>> 11:45:39 AM       all     19.67      0.00      8.25      1.58     70.50
>> 11:45:42 AM       all     13.33      0.00      7.00      1.25     78.42
>> Average:          all     16.92      0.00      7.77      2.47     72.85
>>
>> SQL>
>>
>> So, V$OSSTAT tells me that my CPU resources are 52.9% busy while sar
>> tells me that those very same resources are 70% idle. The "top" monitor
>> confirms that "sar" is right and not Oracle.
>> --http://mgogala.byethost5.com

> 
> V$OSSTAT is cumulative, so wouldn't you have to...
> 
> SQL> declare
>   2    l_secs number;
>   3  begin
>   4    for i in 1..5 loop
>   5      select value into l_secs from v$osstat where stat_name =
> 'USER_TIME';
>   6      dbms_output.put_Line(l_secs);
>   7      dbms_Lock.sleep(3);
>   8    end loop;
>   9  end;
>  10  /
> 30191
> 30200
> 30205
> 30211
> 30216
> 
> PL/SQL procedure successfully completed.
> 
> SQL>

Hmmm, I'll wait for the "METRIC" table to appear. In case someone was wondering, there is no such table in 11.2 ... yet.

-- 
http://mgogala.byethost5.com

Received on Wed Jul 21 2010 - 16:07:18 CDT