Evan Martin wrote:
On 9/8/05, Domas Mituzas Domas.Mituzas@microlink.lt wrote:
Why not cache prerendered versions of all pages?
We have multiple caches. [explanation snipped]
Thanks for the explanation! And with all of that, rendering is still the bottleneck?
Not really, Article::view is only 19% of profiling time, based on the figures I posted. As I was saying, I prefer to discourage the "bottleneck" mindset. An average request is very diffuse. It's meaningful to talk about the bottleneck of some slow functions, but there's no "bottleneck of MediaWiki".
To put it another way, in that profiling run, there were 17983 requests. 10556 (58%) of them were action=view. 8952 (49%) were for the current revision, the other 9% were diffs and old revisions. Of those 8952 requests, about half would have been parser cache hits (current hit ratio is 53%). So only about 23% of requests require the rendering of the current article text. And this is meant to be our bottleneck?
-- Tim Starling