Bernd Helmle wrote:
> --On Dienstag, Juli 21, 2009 16:49:45 -0400 Andrew Dunstan
> <andrew@dunslane.net> wrote:
>
>> You just tested COPY, not pg_dump, right? Some pg_dump numbers would be
>> interesting, both for text and custom formats.
>
> Plain COPY, yes. I planned testing pg_dump for this round of my review
> but ran out of time unfortunately.
>
> The restore might be limited by xlog (didn't realize that the profile
> shows XLogInsert in the top four). I'll try to get some additional
> numbers soon, but this won't happen before thursday.
>
If the table is created by the restore job, either use parallel
pg_restore (-j nn) or use the --single-transaction flag - both will
ensure that the WAL log is avoided.
For plain COPY, get the same effect using:
begin; truncat foo; copy foo ... ; commit;
All this assumes that archive_mode is off.
cheers
andrew