TNHarris wrote:
I've mislaid the previous search times I had, though I guess some are on the mailing list somewhere, so I have no concrete figures. I did run some searches earlier today, the results were generated noticeably faster than before. Search results were being generated in a time comparable with a normal page view. Maybe a little slower than when a page is pulled from cache, but probably a little quicker than when a page is generated from wikitext.These numbers don't mean anything in an absolute sense, of course. This was measured on a desktop computer, and not a very fast one. Like most benchmarks, it only matters for measuring relative performance.
Thats a really good result. Great work. I haven't looked at the code, so maybe this is way off the mark.Would you gain some more improvement by ordering the words and searching with a non-linear method?
e.g. from your list: linux 0.010 towel 0.028 angel 0.082and without knowing other factors (number of results returned, position in word list) those results suggest a potential 3-8 fold improvement.
Cheers, ChrisPS. I checked three five letter words at beginning, middle and end of my index. They each returned exactly one result, no pagenames. #52 0.122 #888 0.152
#1756 0.153Those are complete page timings, which would indicate that the impact of further savings in searching may not be great.
-- DokuWiki mailing list - more info at http://wiki.splitbrain.org/wiki:mailinglist