RE: 10G and UFS - long write times
- From: "Stig Hornuff" <hornuff@xxxxxxxxxxx>
- To: John.Hallas@xxxxxxxxxxxxxxxxx, F.Castillo@xxxxxxxxxxxxx, oracle-l@xxxxxxxxxxxxx
- Date: Thu, 08 Jun 2006 09:23:51 +0000
As I read your description you are basically comparing I/O writes on SAN
disks and local disks.
Is that what you are doing on various Oracle versions?
If it is, then note: Solaris using local disks (ie. in the very same box as
the CPU) will Always perform a write all they way down on disk - verified
writes!!
If disks are remote/attached Solaris "allows" itself to rely on writes being
successfully submittet to the remote I/O-systems (ie. the SAN, T3,
whatever....).
In simple words: Local disk are waaaaayyyyy slower than SAN and remote disk
subsystems.
/S
-----------------------------------------------
True Felix but that is not the point.
I wanted an easy script to recreate a problem
It does not explain why it takes say 30 seconds on 9i and 9 minutes on 10G
on boxes that should be set up the same.
I use the word should carefully as I am assured we have a very conformant
build standard that applies to all servers. I cannot guarantee that myself
and frankly do not have the skills nor the time to do so.
John
_____
From: F.Castillo@xxxxxxxxxxxxx [mailto:F.Castillo@xxxxxxxxxxxxx]
Sent: 08 June 2006 09:19
To: Hallas, John, Tech Dev; oracle-l@xxxxxxxxxxxxx
Subject: AW: 10G and UFS - long write times
Hi
if you're really using this script you're issuing 50000 commits - no wonder
it takes so long. Make the commit at least outside the loop!
Felix
-----Ursprüngliche Nachricht-----
Von: oracle-l-bounce@xxxxxxxxxxxxx [mailto:oracle-l-bounce@xxxxxxxxxxxxx] Im
Auftrag von Hallas, John, Tech Dev
Gesendet: Donnerstag, 8. Juni 2006 10:16
An: oracle-l@xxxxxxxxxxxxx
Betreff: 10G and UFS - long write times
We are seeing big problems when issuing disk writes using 10G (both 10.1 and
10.2). Reads are not an issue
Very difficult to track down the exact combination but so far it looks as if
it is a combination of 10G and local disk mounted using UFS .
V440 Solaris servers could also come into the combination because that is
what I have done a lot of the testing on as that is what is available. I
have now tested on around 10-12 servers with consistent findings.
The test is very simple and quick to do. Searching Metalink and WebIV does
not give any clues. I am about to log a tar but I know I will require all
sorts of Unix information which is quite long-winded for me to get so any
help would be appreciated.
The test
CREATE TABLE test( a number, b CHAR(2000) )tablespace xxxx ;
Set timing on
BEGIN
FOR r IN 1..50000 LOOP
INSERT INTO test (a,b) VALUES (r, 'test');
COMMIT;
END LOOP;
END
On SAN disk and 9i on local UFS mounted disk I am seeing response times from
0:15 to 0:30 seconds
On 10G on San I see the same
On 10G using local UFS disk the time various from 7:00 to 9:00 minutes and
is very repeatable.
I have tried around 10 servers now and the results are consistent.
Whilst we have quite a few 10G installations, not many are on local disk and
the majority are on Sun 440 (2.9 OS) so my results may indicate a UFS
problem when it could be an OS one (or an Oracle one of course). OS writes
to disk are stable at twice as fast on San as local disk and the fact that
90i does not show a problem seems to take away the OS part out of the
equation. But nothing is certain yet.
PS the waits are all IO related - log buffer , log file sync etc
Thanks for any help
John
--
//www.freelists.org/webpage/oracle-l
Other related posts: