Thanks Roger. There is no other activity on this machine. I have run vmstat 5 during all of the different cases. For plain sequential files, Linux is able to almost completely overlap I/O with CPU, so the real time is close to user+sys time. With SQLite, there are often periods where the CPU is stalled waiting on I/O.
I still can't understand how 1 long transaction (no syncs) can be slower than periodic 5-second commits. Any ideas? Jim On 2/4/09, Roger Binns <rog...@rogerbinns.com> wrote: > > Jim Wilcoxson wrote: >> Can anyone shed light on why building a database inside a single >> transaction would be slower than periodically commiting? > > If you look at the user and sys times then each approach is almost the > same. The real times differing so much indicates other activity on the > machine as well as I/O wait. _______________________________________________ sqlite-users mailing list sqlite-users@sqlite.org http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
