Fred Moyer <fred@redhotpenguin.com> writes:
> On Wed, 2004-03-10 at 15:30, Tom Lane wrote:
>> A while, sure, but 2 hours seems excessive to me too.
> If there are no foreign keys or triggers and updating each row is taking
> one drive seek ( approximately 9 ms with the 80 gig IDE drive being used
> here ) then to do 747524 seeks will take 6727716 ms, about 10% less than
> the time of 7628686 ms for the update above. Is this is an accurate
> estimate or are these numbers just coincidence?
Probably coincidence. There's no reason to think that a large UPDATE
would expend one disk seek per updated row on average --- there's enough
buffering between the UPDATE and the drive heads that under normal
circumstances the cost should be lots less.
If I had to bet at this point I'd bet on inefficient foreign-key checks,
but since we haven't seen any schema details that's purely speculation.
regards, tom lane
