Subject: | |
From: | |
Reply To: | |
Date: | Mon, 8 Dec 1997 09:23:31 -0600 |
Content-Type: | text/plain |
Parts/Attachments: |
|
|
At 05:40 PM 12/7/97 -0700, Joe Geiser wrote:
>2/ Tested both updating (using an UPDATE action query) and the
>DELETE/INSERT action queries with the same results as far as the sort item
>was concerned. Of course, the DELETE/INSERT took a heck of a lot longer.
I whole heartedly endorse Joe's findings! The testing/characterzation work
that Wirt and I conducted prior to release of CIUPDATE showed that using
CIUPDATE improves performance from 50% to 800% over DELETE/PUT, depending
upon the number of paths involved. A meaningful portion of this savings
involved the reduction in I/O. That's what we, along with Alfredo and others,
have long been a proponent of using CIUPDATE.
Also, as Wirt had indicated earlier, we spend a lot of time explicitly
testing CIUPDATE with sorted paths, trying to find fault with the
implementation. We devised a number of not-so-obvious test scenarios
yet could not find any situation whereby the ordering of entries was not
properly handled by IMAGE.
However, since that time there have been a lot of other changes to IMAGE.
It may be possible that a problem had been introduced. Before completely
discounting the possibility of a problem, it would be prudent to fully
examine the scenario along with the before/after data to determine if
indeed there currently exists a problem.
If Jeff can contact me privately with more details on his scenario I would
be happy to investigate, as I, along with several others have 12+ years
invested in progressing this enhancement.
/jf
_\\///_
(' o-o ')
___________________________ooOo_( )_OOoo____________________________________
Friday, December 5th
Today, in 1933 - The 21st Amendment took affect.
Tomorrow (Sat), in 1790 - The U.S. Congress met for its first
session in Philadelphia.
___________________________________Oooo_____________________________________
oooO ( )
( ) ) /
\ ( (_/
\_)
|
|
|