On Mon, 2002-11-25 at 23:27, Ron Johnson wrote:
> On Mon, 2002-11-25 at 21:30, Tom Lane wrote:
> > Ron Johnson <ron.l.johnson@cox.net> writes:
> > > On Mon, 2002-11-25 at 18:23, scott.marlowe wrote:
> > >> The next factor that makes for fast inserts of large amounts of data in a
> > >> transaction is MVCC. With Oracle and many other databases, transactions
> > >> are written into a seperate log file, and when you commit, they are
> > >> inserted into the database as one big group. This means you write your
> > >> data twice, once into the transaction log, and once into the database.
> >
> > > You are just deferring the pain. Whereas others must flush from log
> > > to "database files", they do not have to VACUUM or VACUUM ANALYZE.
> >
> > Sure, it's just shuffling the housekeeping work from one place to
> > another. The thing that I like about Postgres' approach is that we
> > put the housekeeping in a background task (VACUUM) rather than in the
> > critical path of foreground transaction commit.
>
> If you have a quiescent point somewhere in the middle of the night...
>
You seem to be implying that running vacuum analyze causes some large
performance issues, but it's just not the case. I run a 24x7 operation,
and I have a few tables that "turn over" within 15 minutes. On these
tables I run vacuum analyze every 5 - 10 minutes and really there is
little/no performance penalty.
Robert Treat
