--- Dennis Cote <[EMAIL PROTECTED]> wrote: > I made a modified version of your test program. It > uses the preferred > prepare/bind/step/finalize call family to execute > the SQL. This way it > only parses the SQL once, and executes it many > times, each time using > different parameter values. I created a similar > program using the > equivalent sqlite version 2 calls. Both programs are > included below.
There is a reason my program is like that: in production, queries will have no correlation and therefore cannot use statement preperation or transactions (I need the "autocommit" mode). > * transactions speed up memory inserts on version 3 > by a factor of 10, > but slow down memory insert on version 2 by about > 6%. I think this behaviour is conter-intuitive for a memory database - there's no need for locking and IMHO (not an expert :) ) these should run almost the same as without transactions. For now, I'll use sqlite2 for my application. Thanks to everyone for interest and feedback! __________________________________________________ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com