Thanks for above comments. I take point about comparing apples and
oranges but test results such as these are the things that management
decisions will be/are being based on so I am trying to make the best
comparison I can.
Regarding SGA I have been doing test with this at 44MB this gave me
hit ratio of .99, I tried raising SGA to 300MB which did give some
improvement in times to between 35-40 secs with hit ratio of .99.
Other details: I am using Solaris DirectIO option which did give
significant improvements especially for redo logs, the test table I
use has 5 cols one of which is PK(index in separate tabspace)
populated from sequence, table in own tabspace < 12MB, redo logs 150
MB on one disk(log buffer 160k), datafile and Rollback segs(extents
20MB) on the other two disks, I do COMMIT after Update completes but I
find all the writing takes place before this, during the Update I can
monitor with iostat and writing starts with start of Update and
finishes with end of job, COMMIT only produces small amount, I guess
Checkpointing, most of writing is to Rollback(from V$Filestat).
I have run the tests on SQL Server installed on my PC (450CPU, single
disk and 128 Mb Ram), which maybe gives a better comparison the
smaller memory forcing it to do some I/O during TX, this gave similar
results(40 secs or so) to Oracle on the Sun server but I'm not sure if
this is being unfair to SQL Server also unrealistic regarding hardware
market. Maybe its that the performance for these tests from both
databases is pretty similar but SQL Server is better at smaller table
caching and SQL Server is on the latest hardware and Oracle on some
fairly old hardware.
Garrick
Received on Fri Jan 04 2002 - 11:23:48 CST
