RE: Curioser and Curiouser

  • From: "Cary Millsap" <cary.millsap@xxxxxxxxxx>
  • To: <oracle-l@xxxxxxxxxxxxx>
  • Date: Fri, 14 Jan 2005 08:13:05 -0600

When I was at Oracle, many, many customers would ask me "privately" which
platform really is the best for running Oracle. The honest answer was that
I've seen Oracle run really well on just about every platform there is, and
I've seen Oracle run really poorly on just about every platform there is.
The number one ingredient in the performance of Oracle is whether there's a
PERSON in the system who has the will and the skill to make it run
efficiently. If you have that, you can run Oracle on anything.

Cary Millsap
Hotsos Enterprises, Ltd.
* Nullius in verba *

Upcoming events:
- Performance Diagnosis 101: 2/23 Houston, 3/16 Salt Lake City
- SQL Optimization 101: 2/7 Dallas
- Hotsos Symposium 2005: March 6-10 Dallas
- Visit for schedule details...

-----Original Message-----
From: oracle-l-bounce@xxxxxxxxxxxxx [mailto:oracle-l-bounce@xxxxxxxxxxxxx]
On Behalf Of Niall Litchfield
Sent: Friday, January 14, 2005 7:18 AM
To: jkstill@xxxxxxxxx
Cc: mgogala@xxxxxxxxxxxxxxxxxxxx; zoran_martic@xxxxxxxxx;
Subject: Re: Curioser and Curiouser

On Wed, 12 Jan 2005 09:00:16 -0800, Jared Still <jkstill@xxxxxxxxx> wrote:
> On Wed, 12 Jan 2005 11:27:26 -0500, Mladen Gogala
> <mgogala@xxxxxxxxxxxxxxxxxxxx> wrote:
> > I can answer that, but being that I am a Linux fan, I don't really want
> > because Win2k (have not tried 2003) was consistently beating FC3 (same
> > machine, just as in Niall's case) by 5%-10%. This orange juice must be
> > too strong for me.

FWIW turning on directIO (I have also setup for async io but here it
is LGWR that is too slow) brought the FC3 score down to 84s which is
slower than winXP, but I'm prepared to believe that this is probably
about as good as it gets for this particular script - and quite
possibly unnoticeable to end users.

> It must be FD2 config.
> I've run the same databases on RH 7.1 and Win2k, with the machines
> being identical ( Dell PowerEdge 2550, 2 gig RAM)
> Well, not quite identical.  The Win2k box has more disks, and was running
> only one database.
> The RH box was running 4 databases, 2 very active.
> And it was still faster than the Win2k.

So in fact not the same at all :). 

Its worth noting that my results are exactly worthless for comparing
linux and windows as yet. One *might* conclude that the lgwr process
on FC3/ext3  is less efficient than the lgwr process on Windows/NTFS,
but that is about it.

I'd actually expect Linux to do better in a controlled test of
identical databases suffering similar external load and a real world
mix of ddl and dml - i.e. something similar to Jared's anecdote.  In
particular someone on this thread mentioned scalability and I *expect*
the mutli-process/shared memory architecture of *nix to do better than
the private memory/multi threading architecture of windows.

Actually I'd suggest one more thing. That in both cases the gains from
tuning far outweigh the innate platfom differences. If this is
consistently the case - and I have to confess that I expect it to be -
then the question one should be asking is not (which is faster linux
or windows) but how well do I think this thing is tuned?

Niall Litchfield
Oracle DBA


Other related posts: