John Bachir wrote: > i've read other posts on this list that say that we can't guess what sqlite > will do with cache.
It uses a simple LRU algorithm to determine which pages to kick out of the page cache first (so at least it's somewhat deterministic). > however, could i be relatively confident that most of the time, it > will prioritize keeping the index in memory before it starts keeping > the data? The page cache does not know what is in the pages. Let's look at a simple example: assume the index has two pages, X and Y, which each point to records in three data pages: X -> A,B,C; Y -> D,E,F The order in which the pages would be used is this: X A X B X C Y D Y E Y F For LRU, the last usage matters, so the LRU list will look like this: A B X C D E Y F So the data pages _will_ crowd out the index pages (especially when there are much fewer index then data pages ). > ideally it would always keep the entire index in memory and never > cache the data. > > if i can't more or less depend on this, then sqlite probably won't > work for my application. You could write your own page cache implementation that wraps the original one but never throws out certain pages ... Regards, Clemens _______________________________________________ sqlite-users mailing list sqlite-users@sqlite.org http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
