Re: PEAR Algorithms/Containers

On Thu, May 18, 2006 8:57 am, tedd wrote: > However, binary-trees are for searching and I use MySQL for that. I > don't know specifically how MySQL preforms searches. I had hoped that > somewhere one could set a bit and change the algorithms, but I > haven't been able to find any references that allow that. You really would need to discuss this at length on a MySQL forum/list, but basically, the first thing you have to look at is the indeces, if any, on the tables in question. If you have an index, it could be a B-tree, or a hash, or any other number of low-level indexed structures which more-or-less imply the algorithm to be used in a search/sort. The second thing to examine is MySQL's estimate of the number and distribution of records to be loaded/examined/returned. If the result set to be loaded/examine/returned is relatively high number and widely distributed over hard drive "pages" then MySQL "knows" it has to do a full table scan anyway, and goes ahead and does that, I think. All of the above is a gross over-simplification of something that MySQL has put a zillion man-hours into solving -- which is why a database is so GOOD at this sort of thing. Of course, one doesn't always have a choice... For example, one might be faced with a dog-slow crappy Google adwords API feed that can only return keyword traffic estimates for specific bids, and then one would be forced to hit that dog-slow crappy Google adwords API feed in a binary search algorithm just to have any hope at all of completing one's task in some reasonable time-frame... Not that I've ever faced this problem, oh no. :-( To me, the added complexity of a class to handle something as simple as binary search, hash-search, or even a B-tree or order N, is just overhead and complexity of integration that I don't need. When I need that kind of algorithm, I do not find it difficult nor tedious to just code it from scratch. But I can see where others would find it a real boon, so I suspect it would be quite useful for many to find in PEAR or some reasonable PHP code library. What might be MORE useful, perhaps, would be an extension in C to add some of these algorithms to the language somehow -- Since that would maybe include a performance boost in the algorithm itself, assuming you could get the data/structures into and out of the C code in some reasonable way. YMMV -- Like Music? http://l-i-e.com/artists.htm