2009-10-01 19:37:34 UTC
Hi,
The option for keyandthink is by default set to "on or off", the reason is because the values are the same as the ones in the TPC-C specification. There is nothing to stop you experimenting with changing it though in the driver script, just one of the advantages of Open Source! I would say though that to use the 'official values' you do need a huge database with hundreds of thousands of warehouses to create any significant load eg the pause time before a new order is 17 seconds.
To do what you want to turning keyandthink off to compare CPU performance is fine and the way I would recommend to start running tests. Yes by thread I mean Vuser - I am just using the technical term as in reality each Vuser is an operating system thread and not a process but each is an independent user you can see logged in the database in v$session. So in terms of your tests when keyandthink is turned off the workload is intensive so you should definitely see 8 vusers generating more of a load than 1 Vuser! You should be able to graph the results (generated in an AWR report test) in a spreadsheet and see the transaction rate increase in proportion to the Vuser rate. This way you should be able to find the number of users you need to generate your steady 50% CPU load and run your hour long test. If you can't get the CPU rate up then you have a bottleneck elsewhere in memory or I/O. To find it use the AWR reports and refer to the link given previously in this thread, CPU must be the top event in the top 5 timed events if it is not then it is stating the obvious that you are not comparing CPU performance. Feel free to post the top 5 timed events here
Cheers,
Steve
