How close to this is PG's COPY? I get surprisingly good results using COPY with jdbc on smallish systems (now if that
patchwould make into the mainstream PG jdbc support!) I think COPY has a bit more overhead than what a Bulkload
featuremay have, but I suspect it's not that much more.
|| Steve, I do not know. But am reading the docs now, and should figure it out. Ask
me later if you remember. Oracle's "direct path" is a way of just slamming blocks
filled with rows into the table, above the high water mark. It sidesteps freelist
management and all manner of intrablock issues. There is a "payback", but the benefits
far far outweigh the costs.
> Now...if you ask me "can this work without Power5 and Hitachi SAN?" my
> answer is..you give me a top end Dell and SCSI III on 15K disks and
> I'll likely easily match it, yea.
>
> I'd love to see PG get into this range..i am a big fan of PG (just a
> rank newbie) but I gotta think the underlying code to do this has to
> be not-too-complex.....
It may not be that far off if you can use COPY instead of INSERT. But comparing Bulkload to INSERT is a bit
apples<->orangish.
|| Oh! I see! I had no idea I was doing that! Thanks for pointing it out clearly to me. Yea, I would
say a full transactional INSERT of 5K rows/sec into an indexed-table is a near-mythology without significant
caveats (parallelized, deferred buffering, etc.)
--
Steve Wampler -- swampler@noao.edu
The gods that smiled on your birth are now laughing out loud.