Skip to Main Content
  • Questions
  • Performance differences between UNIX, Linux, NT, and 2000.

Breadcrumb

Question and Answer

Tom Kyte

Thanks for the question, Debra.

Asked: December 12, 2001 - 9:01 am UTC

Last updated: April 05, 2007 - 11:01 am UTC

Version: 8i

Viewed 1000+ times

You Asked

I am trying to find documentation on case studies comparing peformance issues when running Oracle on various operating systems.

Thanks,
Debra

and Tom said...

We have no such document. We are operating system agnostic, we exploit features of each. We can more than do the job on each. Your choice of OS should be made in favor of what features you want, what skills you have, what you already use.

I'm pretty sure that given any system, you can always make it go 10% faster. Tuning in itself is a never ending goal unless you have some metrics to meet. What you really need to do is figure out what you want to do performance wise, then size the system you would need with each OS (typically done via benchmarking) and see which one is the best fit.

Rating

  (5 ratings)

Is this answer out of date? If it is, please let us know via a Comment

Comments

But...

A reader, December 12, 2001 - 2:50 pm UTC

Interesting question, but there must be some guidelines you can give. e.g. a really large application or data warehouse probably isn't suitable on an Intel platform and hence NT or 2000 (assuming NT/2000 is generally limited to Intel). Is this a myth?

Tom Kyte
December 12, 2001 - 3:39 pm UTC

You can do very large databases on win nt/2k -- if you know what you are doing. You can do very large databases on unix -- if you know what you are doing.

As I said -- it depends on what you are comfortable with, what you know, and how much you really know it. We have very large databases on both -- its more of a hunch on my part, but if pressed i would say most of the really large ones are Unix based.


Chapter 2 of your book and benchmarking

Sam, December 12, 2001 - 5:40 pm UTC

In chapter 2 of your book, talking about different kind of databases, you talk about 'BENCHMARKING' , I did not understand.

Can you let us know what is the concept of benchmarking.

Tom Kyte
December 13, 2001 - 8:30 am UTC

from the dictionary:

Main Entry: bench·mark·ing

Function: noun
Date: 1976

: the study of a competitor's product or business practices in order to improve the performance of one's own company---------------------------------------------

one way to benchmark is to implement a process using the best practices for each product -- run them -- and see which performs adequately for your environment.

another way to benchmark is using a tool like statspack -- to gather and retain historical performance statistics so that over time you can see how you are doing, you can benchmark your current performance against past performance.






Performance benchmarking

Pushparaj Arulappan, August 28, 2003 - 4:49 pm UTC

Tom,

We are planning to migrate to a new HP Server which is
called SUPERDOME. It has been told that the SUPERDOME
server has the very powerful processors. Currently we have
HP V25000, CPU=16 @ 440 MHZ, Memory=32 GB)
and the new server would be CPU=12 @ 875 MHZ, mem=32 GB.

We have 8 Oracle9i instances running in the current server
and the same will be migrated to the new server.

Is there any performance benchmark that I can do in the old server and repeat the same test on the new server after the migration to verify if we get any performance improvement
on the new server.

Thanks
Pushparaj

Tom Kyte
August 29, 2003 - 8:39 am UTC

8 instances is of course 7 too many....


you could run a tpc-c on the old/new servers -- but that would just tell you how fast/slow your hardware can run a tpc-c.

it is really hard to benchmark an 8 instance server -- you cannot benchmark individual instances as they all interfere with eachother. It is in my opinion a really bad idea to have many instances on a server, no way to manage resources, to control things. I'd rather have 8 2cpu machines or 1 16 cpu machine with a single instance.

Benchmarking my database before switch over

Orakle_Lover, January 18, 2006 - 11:37 am UTC

Hi Tom,

We are going on to new hardware this weekend. could you suggest some ways to benchmark by database, so that I can compare the database performance before going to the new hardware and after.

I am taking the statspack report and using the oraperf.com. But I am in the search of some other benchmarks too.

Database Version is 9.2.0.6
Os is AIX 5.2
Database size 800Mg


Tom Kyte
January 19, 2006 - 7:55 am UTC

collect the statspack every 30 to 60 minutes during your normal load for the rest of this week.

do same next week.

You'll now have metrics to compare before and after. query elapsed time, cpu time, transaction rates, IO's, etc...

Naveen samala, April 05, 2007 - 7:06 am UTC

Hi Tom
Excellemt , can you please eloborate on this how to look and how to compare? It will be very much useful for us.


Tom Kyte
April 05, 2007 - 11:01 am UTC

umm, just run two reports and look at the numbers?

.... query elapsed time, cpu time, transaction
rates, IO's, etc... ....