On May 7, 2009, at 3:27 AM, Pavel Ivanov wrote: > Hi, all! > > Recently I've discovered that my application works faster if it uses > journal_mode = PERSIST instead of MEMORY. After running under > callgrind I've found that with journal_mode = MEMORY application > spends a lot of time inside malloc() which is called from > memjrnlWrite(). Is there a way to override this? I mean, I'd be happy > to implement my own vfs for memory journal to avoid such intensive use > of malloc(), but it seams that there's no way to do that. Am I wrong > and there is actually a way to do it?
At the top of file memjournal.c there is a #define called "JOURNAL_CHUNKSIZE" that sets the size of the allocations used to store the in-memory journal. Currently it is defined to (just under) 1024. Maybe you could try setting it to a larger value and recompiling. > Now I want to add into application work with in-memory database. For > this kind of databases there's only 2 possible values for journal_mode > - MEMORY and OFF. I'm afraid that with MEMORY I'll get the same > negative performance impact, so I'm leaning to OFF. But if I have > journal turned off, make insert into the table and it fails because of > unique constraint will it be able to rollback this insert statement? > And more general: has journal any meaning at all when all my > transactions contain only one auto-committing statement? In this case it will only matter if you have statements that affect more than one row that could hit a constraint. i.e. an UPDATE or an INSERT INTO ... SELECT statement. Dan. _______________________________________________ sqlite-users mailing list [email protected] http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
