FYI - my Oracle on Linux/Windows comparison results..

I finally managed to snare two identical machines for a more or less
"apples to apples" comparison, more for my own curiousity than anything.
Anyway, machines are generic type clones with new ASUS P4PE 3Ghz motherboards and 750MB ram each, windows 2000 pro SP4 on one and RH linux 9 on the other. Installed 9.2 on each, patched to 9.2.0.5 and created clone of a production system on each, allocating same shared pool etc.. for both. Both machines have a single drive only (at this point)

After refreshing each Db from a consistent export, I ran some queries on   one of the bigger tables, which is 6+ million rows, made up of several number columns and one varchar2. For query #1 I selected sum, average, count,and calculated average of one column grouped by another for 2 groups. Results were in 17-18 seconds for windows machine and 11-11.5 seconds for linux, query was ran 6 times each.

I then created a copy of the same table with no indexes (create table
"copy" as select * from "original"), this took 3:37:02 on windows and
1:05:00 on linux. Then re-ran same query, results were 1:06:07 -1:07:09 for windows machine and 00:11:03-00:15:07 for linux, again, 6 times each.

I then created 1000 row non indexed table (range_table) on each with a low-high range of numbers and a range number column:

start_r end_r r_number
======= ===== ========

200	204	1
205	209	2
210	214	3.....

I then ran a query that summed and averaged the same column in the main table, joined it with the range_table and grouped by the r_number col in the range_table. Results were consistently around 7:34 on windows and 1:20 on linux. I then opened one more session on each and ran both queries concurrently on each machine, the linux machine took 68% more time to return each result and windows about 99% more time.

One thing I did notice was that the windows machine's CPU was between 8 and 15% usage while queries were running except for a short spike to about 40% near the end and linux was always close to 100% from the time the query was executed. Changing the priority for the oracle process in the windows machine seemed to have little, if any effect.

I wasnt sure what to expect from this comparison, but I was not expecting such a huge difference, I have a lot more testing I want to do, including turning off the GUI totally on the linux machine and trying different filesystems, but thought I'd post these results for now and see of anyone else on the group has done similar comparisons, and may perhaps have some suggestions/comments on ways to enhance or improve comparison tests.

cheers Received on Mon May 10 2004 - 15:45:03 CDT