Benchmarks

This message was sent from: Suggestions and Proposals
http://forums.oscommerce.com/viewtopic.php?p=204876#204876
----------------------------------------------------------------

Well since there's new code, I had to do some new benchmarks:

The results below are for a stock install of osC on a clean database.
Siege runs concurrent requests to the site, with urls randomly picked from a list of URL's. 
All tests were run for 120 seconds.
Before each set of runs both apache and mysql were stopped and started again.
The tests were run on the demo server from the previous post.

[i]Attention:[/i] These times are [b]overall page performance results[/b], not just the improved parts, so this gives a pretty good idea of the overall impact of the improvements on osC performance.


[i]With SHOW_COUNTS = 'true'[/i]
Both connection rate and response times improved consistently by [b]13-15%[/b].
I was surprised by the numbers here, I didn't expect it to show so well on a small database.

[i]With SHOW_COUNTS = 'false'[/i]
Both connection rate and response times improved marginally consistently [b]2-3%[/b].
Even with a larger database the difference will not be too great here, there are simply less queries to be done away with.

And because a picture says more than a thousand words, these numbers looks something like this with SHOW_COUNTS = 'true':

[img]http://www.cryp2nite.nl/osC_ori_vs_new_siege_bench.png[/img]


[b]Notes[/b]
1. I tried testing this on  a development machine, but was unable to get results for 50 connections for the old code because the webserver went unresponsive.
2. I still would like to see how this performs on a database with more catageories and products, because that is were I am expecting the real peformance gain.
3. The layout code isn't very good wrt performance, I'm convinced I can gain another couple percentage points there due to unnesessary copying of and looping over large arrays.


Hope you guys like it,
Rob