Re: DB time Less Than Elapsed Time - is there a performance problem? or is the report interval incorrect?

From: David Roberts <>
Date: Sun, 19 Sep 2010 12:37:23 +0100
Message-ID: <>

I'll admit I'm more of a DBA 0.9 type person, but my approach would be to always start at the OS level before looking inside Oracle.

So I'd run sar 10 10 and sar -q 10 10 (ten samples at 10 second intervals) and then I'd also look at the historical sar and sar -q data over the previous few hours.

While this may just re-in-force your initial understanding that the CPU isn't busy, it might show a significant variation historically in terms of the demand on the CPU.

It might also indicate high I/O, but at least it should give a very high level idea as to where in the system there may be a bottleneck.

I'm not personally familiar (in terms of hands on) with RAC, but the nodes seem to be performing somewhat different levels/types of workload. I would be interested if this was by design, ie. certain types of users are pointed at different nodes, or if this pattern was just a blip, and the workload on each node was usually similar.


On Sun, Sep 19, 2010 at 4:22 AM, Greg Rahn <> wrote:

> Look at the "Host CPU" section to see how busy the system is from an OS
> perspective. This system barely looks busy enough to have a significant
> problem, but there could be something lurking at a statement level.
> On Sat, Sep 18, 2010 at 3:28 PM, Oracle Dba Wannabe <
>> wrote:
>> Hi All,
>> I've been given two awr reports for a 2 node, RAC instance
>> running on AIX and told to determine what the performance problem is with
>> the system. Looking through the reports I can't say that I can see anything
>> wrong with this system - at least based on the awrs I have been given (a few
>> extracts from the reports given below). What I have noticed is that for both
>> nodes the db time is considerably less than the elapsed time of the
>> report - is this enough to deduce that this system is actually not busy and
>> not performing any sort of work?. That said, the 'transactions per second'
>> statistic is at 0.6 and 1.6 (for each node), additionally , logical reads,
>> physical reads appear to be quite low.
>> Or perhaps one of the following possible:
>> (a) Since DB time excludes a few idle events - such as the sql*net related
>> events that perhaps time is being spent there are not being recorded?
>> Therefore db time is less than elapsed time and gives the illusion that the
>> system is not busy?
>> (b) That perhaps the performance problem is short lived - lasts for a
>> duration less than the elapsed time - therefore is not evident from these
>> two awr reports
>> Thanks in advance for any pointers or help.
> --
> Regards,
> Greg Rahn

Received on Sun Sep 19 2010 - 06:37:23 CDT

Original text of this message