Some comments on the last indexer update. Indexing was improved by dumping the piecemeal reading of i*.idx and just loading the file all at once. This made a huge difference. Times dropped as much as 60% what it was before. I also got a better sense of how the functions perform. A lot of the noise in my first tests was caused by having to hit the disk so often. Without that, I can clearly see that idx_getPageWords doesn't scale that well. It's not very severe, but that's where any improvements should be focused. Searching is also improved a bit, but not as much as I expected. Statistically insignificant, really. I tried Chris's suggestion to group same-sized words when reading the index. I didn't see as much of an affect from that. If there is an improvement, it may be offset by the extra work that has to be done. The simpler algorithm is probably helped by the disk cache, so there's not as much penalty for re-opening files. But I don't entirely trust my home machine to give reliable results for this. So there are two alternative functions for retrieving search words from the index. You can choose one or the other from the configuration manager ($conf['test_indexer']) Set to 0 for classic searches, 1 to try the new sorted search. This is a bitfield in case I encounter something else to test. -- tom telliamed@xxxxxxxxxxx -- http://www.fastmail.fm - A fast, anti-spam email service. -- DokuWiki mailing list - more info at http://wiki.splitbrain.org/wiki:mailinglist