I spent some time working at a consulting firm where we got to travel around and work on lots of different environments. One thing that really surprised me the most was the lack of quality control for the consultant job title in the industry as a whole. There are some people who know a lot, and some who actually give rather poor advice. Generally they’re all rather confident in themselves.

Interesting thing: even after I realized this, I still caught myself thinking that i was very smart and important just because of my job title. It’s a pride thing, like you said. Funny how you can see it in others, dislike it in yourself, and still have the need to be frequently reminded that the title in itself does not bestow any importance.

It’s good for me to frequently be reminded how lots of people know more than me. Helps me remember to never be embarrased of any question.

I would expect calibrate_io to give you information that was relevant to 10g also. Admitted, different versions of the database/ASM do things differently, but the whole world hasn’t changed, so I’m sure it would be reasonable information.

From an I/O alone perspective, ORION would be much better for you as it can test the impact of combined OLTP/DSS loads on storage performance. Obviously, this doesn’t bring in your CPU issues though.

In your situation, I would probably be more interested in a whole system load test. You could use something like Swingbench (http://dominicgiles.com/swingbench.html) to test that. It comes with some basic benchmarks, but you can also write your own that hook into your application.

Finally, Oracle provide a neat feature in 11g called Database Replay (http://www.oracle-base.com/articles/11g/DatabaseReplay_11gR1.php). You can capture a workload before the disks are changed and replay it exactly as it happened after the disks are changed and monitor the affects. You can capture from 10g, but replay is 11g only, so unless you do a test replay before SSD and after SSD you would be comparing applies with oranges.

Cheers

Tim…

I really appreciate your overview, I just found out I will be evaluating an SSD without near enough time to come up with a proper test harness. I have a mixed OLTP DSS system, and most of the performance issues have been cpu thrashing by the DSS stuff wailing on the SGA. So my initial reaction was hmmmm, if all the OLTP stuff that is performing properly goes faster with an SSD, that will perhaps sink the cpus because of all the other things… I have few I/O problems, even with a low end mostly RAID-5.

So the question is, will the 11g calibrate give answers applicable to 10g? Of course I’m on hp-ux, which is a redheaded stepchild these days, especially where async is concerned. And the device is likely to get other non-oracle VM kind of stuff too (which I expect will perform better in simple-minded testing and provide the impetus to dump the RAID).

If I had to prove everything to make it real, we’d all be a figment of my imagination. But it would probably make a good, er, sequel to the Matrix.