Steve Dirickson wrote:

>Oh, I don't know....How about because I like doing serial reads of detail
>sets? :-)
>
>Seriously now, how does the overhead of a DBGET mode 2 of an automatic
>master plus a DBFIND to the detail set plus a DBGET mode 5 of the detail
>set record compare to a DBGET mode 2 of the detail set?  Is the
>(technically) more complicated DBGET set #1/DBFIND-DBGET set #2
>better/quicker/faster
>than the "cleaner" (or at least clearer) straight forward DBGET set #2?
>
>Inquiring minds want to know!>>

I haven't done this test in about three years.  However, my experience is
that each DB call, regardless of the type are similar in speed on the
larger machines.  Of course, this will depend on whether you are using TPI
or Shareplex or etc., but generally, it's about 1 or 2 millisecond per
call.  But in any case, the real question is "How many DB calls do you have
to make to get your records?"  (DBDRIVER is an excellent method for getting
these numbers from your own machine.)

If a serial read of an automatic master, a dbfind of the detail set and a
chain read from there results in 10,000 I/Os versus a serial read of the
detail set resulting in 100,000 reads, then generally reading the automatic
serially is the best answer. (Yes, pre-fetching, large memory and other
items can be used to bring these items even closer together).

The real problem is that with each data set and with all the available
search products available today, one solution could be very slow on some
sets and very fast on others.  Because of the management issues of making
decisions are what method to use when, most customers that I work with
chose one method that is the "best" for their environment.  They may decide
that they will use IMAGE B-trees or a TPI produce all the time.  Or they
may decide they will use SUPRTOOL all the time, or Steve's method of
retrieval, or just a serial read of the data set.  The key is that the
"expected" performance of one of these methods will be faster than another,
and the company is willing to accept somewhat worse performance in a few
cases than to use 6 different methods throughout their environment.

What's "best" - using multiple methods or choose a single method?  :) :)
 Oh, boy, now you're into an area such as religion.  I'm just showing some
of the many options. :)

LB