Re: interpretation of V$SESSION_EVENT statistics

We really need the values for read times, (it is just possible that you have a spuriously fast sequential read time, rather than a slow scattered read time).

Your init.ora says that your scattered read is trying to do 64K - which tends to correspond to the time taken to do half a rev of a disc i.e. a few extra milliseconds. In practice, however, I tend to find that a 64K read often tends to take around 50% longer than a single read, so your figures look highly suspicious.

The obvious thing to look at is the striping - is a 64K read actually being asked to pick up 16K from each of 4 separate discs ? Queuing theory being what it is, this would be consistent with your 5-6 times time dilation.

--

Jonathan Lewis
Yet another Oracle-related web site: http://www.jlcomp.demon.co.uk

Vsevolod Afanassiev wrote in message
<7ria58$65m2_at_xlprod01.westpac.com.au>...
>When looking at results from v$SESSION_EVENT, I found that
>the value of Average_Wait for "db file scattered read" is 5-6 times
>higher than the Average_Wait for "db file sequential read".
>My understanding is that "db file scattered read" corresponds to full table
>scans
>and "db file sequential read" corresponds to index scans/tables accesses
via
>ROWID.
>Does it really mean that full table scans are so slow?
>
>Thanks,
>Sev
>
>Some info on the system:
>Oracle 7.3.2 on HP-UX 10.20
>HP 9000 K Series, 10 CPUs, 2GB RAM
>300GB database on raw devices (EMC frames)
>db_block_size=8K, db_file_multiblock_read_count=8
>We are running Oracle Financials (GL, AP, PO, FA),
>quite often CPU usage is at 100% (from sar -u)
>
>
Received on Mon Sep 13 1999 - 04:14:28 CDT