Ian Struble wrote: > And just to throw one more wrench into the works. You could load up only > the most popular data at startup and let the rest of the data get loaded > on a cache miss. > > That is one technique that we have used for some customer session > servers. It allowed each server to start up in well under a minute > instead of in 15-30 minutes while pegging the DB. The 15-30 minutes was > when we were dealing with ~5mil total entries and I would hate to see it > now that the size of the table has doubled. Now we just need to do some > batch processing to determine what subset gets loaded at startup.
You could also just dump the whole thing into a Berkeley DB file every now and then. - Perrin