On 24 Nov 2009, at 9:17pm, Pavel Ivanov wrote: > Indeed, it's weird. And I've just realized that if we have two > simultaneous write transactions they both have to write their own > journal whenever they wish to write something to disk. SQLite database > cannot have two different journal files, so it should serialize > transactions whenever they want to actually write something to the > file.
Naturally. We all know that you have only one data structure here. You cannot have two threads/processes/applications actually change the database file at once: they'd find inconsistencies where TABLE data had already been changed but INDEXes had not been updated. You can't have them both looking for a new blank page to write in and accidentally choosing the same page. You need to have one do its work and then the other. This is why a lot of the questions here about attempts to speed things up by using multiple threads/processes/databases are pointless. When it comes down to it you do still have just one computer updating one hard disk. Whatever work has to be done does have to be done in the end. All you're doing is changing the order things are done in. Simon. _______________________________________________ sqlite-users mailing list sqlite-users@sqlite.org http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
