Does Monotone spend most of its database time in a few long running queries or thousands of sub-second queries?
Could you point us to an example monotone database tarball and give some actual examples of speed-deficient queries? When working with large SQLite databases with blobs I generally find I get much greater speed when I move expensive data manipulation from the application layer into a custom SQLite function. I don't know whether anything in Monotone lends itself to this optimization. --- Nathaniel Smith <[EMAIL PROTECTED]> wrote: > In its current implementation in monotone, this algorithm seems to be > seek-bound. Some profiling shows that there are cases where we're > spending more time blocked waiting for read()s for the db, than it > takes to read the entire db 5 times over. This makes sense. We're > basically doing random reads scattered all over the file, so the OS > has no way to do any sort of readahead, which means we're probably > hitting disk seek latency on every read, and like usual these days, > latency trumps bandwidth. __________________________________________________ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com
