Archives of the TeradataForum

Message Posted: Mon, 25 Aug 2003 @ 23:37:57 GMT

	<Prev	Next>		<<First	<Prev	Next>	Last>>

Tom,

The PAC asked for sites to study hash join in 2000, and we signed up. We used the recommended htmemalloc of 1% on a 36-node 5150 with 2Gig of memory/node that's pretty heavily used. It improved run times on some queries a lot, so we're still running it. We've seen a few problems with hash join, but we've also had some with other kinds of joins.

New systems and new features come with risks, which is why we test new releases thoroughly before we put them in production. I've always wondered why this particular system feature has an on/off switch in available in dbscontrol. If NCR had simply incorporated hash join into a system release (without the switch) and treated it like any other join type, we'd have long since wrung all the bugs out of it.

Was there a crisis of confidence in the development organization over the feature? If so, why is it in the system at all? Did a certain large customer lean on them to put the switch in place? We may never know...

/dave "htmemalloc=1" hough

	<Prev	Next>		<<First	<Prev	Next>	Last>>