|
From a general perspective, I have to agree with
Remco. If hints are over-used, they may be "optimal" only for a specific
release of Oracle and/or a narrow range of data volumes and/or in a specific
configuration. Often, they eliminate any other tuning possibilities -
create a new index and the application ignores it, the hint is great for 1000
records but is abysmal for 10,000,000, or partition the data to improve
performance and the app still can't take advantage, etc. This is not to say that hints shouldn't be used, but other
factors - like the level of your influence over the developers to change them as
need arises - weigh in the determination.
As an example, I once spent over six months trying
to get an out-sourced application changed to remove a few very harmful
hints. The contract developers had determined, on their small test
system, that forcing an access-path sort by use of an otherwise wildly
inappropriate index hint and a stopkey was more efficient than using an order by
clause. In production, the reverse was true - the hint caused each
submission of one of these statements CPU utilization of 72x and I/O
of 6200x of what the order by did. This was for a query that was
submitted by their application 50,000 - 100,000 times a day! Guess
who was constantly called upon to "tune the database" to fix this
nightmare...
-Don Granaman
[OraSaurus - Honk if you remember
UFI!]
----- Original Message -----
|
Title: RE: perplexing plan?
- perplexing plan? Doug C
- RE: perplexing plan? Koivu, Lisa
- RE: perplexing plan? Daemen, Remco
- RE: perplexing plan? Christopher Spence
- Re: perplexing plan? Doug C
- RE: perplexing plan? Koivu, Lisa
- Re: perplexing plan? Jonathan Lewis
- RE: perplexing plan? Don Granaman
- RE: perplexing plan? Koivu, Lisa
- RE: perplexing plan? Connor McDonald
