Re: A new proof of the superiority of set oriented approaches: numerical/time serie linear interpolation
Date: 29 Apr 2007 09:13:06 -0700
Message-ID: <1177863186.258396.136950_at_y5g2000hsa.googlegroups.com>
On 29 avr, 16:21, "Brian Selzer" <b..._at_selzer-software.com> wrote:
> I would have done it differently. I would have used a cursor to generate
[Snipped]
I respectfully think you miss the point of the entire thread.
Thinking procedurally (use of cursors) is inherently inferior to set
oriented approaches independently of a specific implementation and
regardless of how one defines performance as response time, ability to
run well in concurrent query environment and so forth.
> I would bet the process would run much faster than your solution. In rare
> cases, cursors can actually increase performance.
I don't think one can establish that for a certainty.
If you ask me I tend to think that even in direct image systems, set
oriented are superior because they give less time outs in highly
concurrent environment. I had the opportunity to take away several
hundred cursors based processes that would run faster on development
environments but would totally time out in heavy load environment. I
have not encoutered so far any cursor based solution that was not
imposed by a poor design. And I have not found any satifying cursor
based solution. Often the use of cursor and (procedural) dictates
unacceptable tarde off where evrything is given up for reponse time
sake...
One should keep in mind that response time is only one of several
criterias that may define performance. Defining performance only
according to response time is simply a consequence of using poorly
designed direct image systems.
[Snipped implementation specific diret image internals]
> I generally don't advocate the use of cursors, but sometimes the optimizer
> just isn't smart enough to generate what I know to be a minimum plan, so in
> those cases I use them.
Speaking of optimizers in the case of direct image systems is like
asking a cow to run a race. Only a TRDBMS that would correctly
separate logicala and physical could inherentlyy allow optimization.
Regards... Received on Sun Apr 29 2007 - 18:13:06 CEST