Re: db file sequential/scattered read - physical or logical io or both?
- From: Martin Berger <martin.a.berger@xxxxxxxxx>
- To: oracledbawannabe@xxxxxxxxx
- Date: Mon, 14 Jun 2010 06:34:16 +0200
Oracle calls sequential and scattered reads 'physical' as these calls
to get blocks leave the Oracle program. Oracle cannot know anything
about the 'external' layers, like filesystem cache or even cache
within a storage array or disk. You can call it a description which
code path within Oracle was used.
Your distinction with the limit of 5 ms is artificial and must be set
up regarding the system your Instance is running on. It might be right
on your system. With such a distinction you can only say something
like 'more than x% of all answers of 5 ms or less are from a cache'.
Keep this in mind if you compare the numbers to real counts e.g. on
filesystem cache hit.
You did not tell us anything about your OS, Hardware and where the
datafiles / Blocks reside. So it's hard to tell if yur lmits of 5 and
10 ms are reasonable.
best regards,
Martin
I suppose what I'm trying to say here is that if the same wait event
is issued regardless of a physical or logical io request, how can I
determine if the io subsystem is returning blocks at s reasonable
service time. Assuming an idle system other than the report, would
times under 5ms be cache hits and times over be physical. Further
anything over 10ms would be a bad service time ? appreciate any
input on this
Thanks
On Mon, 14 Jun 2010 06:31 SGT Oracle Dba Wannabe wrote:
Hi,
So I have an idle system, where I run a sql report. The report
takes an hour to run. I look at the awr report (30 min interval),
and see the io wait events for sequential and scattered reads. The
first has an avg wait time of 7ms the second 10ms. These waits as I
understand it are physical io requests - correct? The p1 and p2
parameters point to file and block numbers so I guess that makes
sense. Anyway I rerun the same report, look at the new awr and now
see the same wait events, only with much smaller wait times. Which
means data was read from cache - if that's the case why are the
same wait events issued? it seems a bit confusing that way.
Thanks
--
http://www.freelists.org/webpage/oracle-l
Other related posts: