On Fri, 15 May 2009 09:45 -0400, "Aryeh Gregor" Simetrical+wikilist@gmail.com wrote:
On Fri, May 15, 2009 at 6:48 AM, Karun Dambiec karun@fastmail.fm wrote:
From looking further into ADODB, it appears that the performance hit from using it could be significant for projects like Wikipedia (14% or so from the statistics on the ADODB site).
14% over what? For trivial queries that take 0.1ms to execute anyway? Not for long complicated ones, I bet. And I'm awfully sure it's not 14% for all of MediaWiki, including the parser and all. I doubt it has more overhead than our current solution -- probably much less, since it does much less. But the overhead is going to be negligible if you run a sane number of queries per request, unless this has way more features than I think.
Some of ADODBs own benchmarks have revealed that it is not necessarily too fast. If there is no accelerator used there is a significant speed decrease in comparison to using MySQL directly. I assume the same would likely apply for other database.
Benchmark 2
The first benchmark is a synthetic one that does not measure live performance. This benchmark tries to be more realistic, measuring HTTP requests/second. In this test, we select and display 82 rows from the products table once per page request.
...
Two runs were taken and averaged for each test. Higher values are better. All measurements in pages/second.
With No Accelerator Accelerator MySQL 83.53 81.35 ADOdb 61.19 21.33 PEAR DB 52.85 25.26