| T.R | Title | User | Personal
Name | Date | Lines |
|---|
| 441.1 | | DPDMAI::DAVISGB | Gil Davis DTN 554-7245 | Mon Oct 02 1989 17:57 | 20 |
| I can't speak about the Ingres vs Rdb, but I can discuss another vendor
versus Rdb as related to hashing. Informix has a brochure that goes
into the evils of Hashing and the benefits of B-tree. When I met them
last Spring, they made a big deal of this is... I asked 'why the big
deal over B-tree and why do you bash hashing as an indexing method?
"Because we don't offer hashing..." he replied. So, the answer is that
the customer should use the BEST solution for the application. I would
guess that either could be shown to be faster if used in the proper
case.
As for Ingres, I've heard many times from our own analysis, from a
former Ingres SDales Rep, and lately from one of their own technical
support people (!) that Rdb is the superior solution on a VMS
platform.. Superior technically, in features and in performance.
Hope this helps..
Gil
|
| 441.2 | Do the disks have to spin clockwise? | WIBBIN::NOYCE | Bill Noyce, FORTRAN/PARALLEL | Mon Oct 02 1989 19:33 | 13 |
| Do you have the time to run the benchmark both ways?
My attitude is, if nothing in the benchmark would benefit from using
B-trees over hashing, then either
- the customer doesn't really need B-trees, or
- the benchmark isn't representative of the true application.
In either case, you have a good reason to go back to the customer
and question the insistence on B-trees. If there's something that
makes them think they need B-trees, they should put it in the benchmark
(and then let you decide how to implement it). Having the different
performance numbers in hand might make it easier to convince the
customer to reexamine their assumptions.
|
| 441.3 | Hashing should be better I guess | ZPOV01::JEFFREYCHOY | Life goes on without me | Tue Oct 03 1989 07:59 | 24 |
|
I have spoken to the customer questioning into why B-trees is used intead
of hashing.
The reply was that they taught B-tree is easier to implement.
I then explained that based on their requirement, Hashing is most
appropriate and Rdb hashing is more superior. They do not mind at all
if I can submit both B-tree and Hashing results of the benchmark.
However, my only limitation is time. Anyway, my strategy is to first
comply with using B-trees for meeting the dateline but then submit another
set of results based on Hashing.(if it is better)
I have used both hashing and B-tree in a previous benchmark but the
difference is hardly noticeable. This perhaps is due to the small size
of the database being used and therefore cannot take advantage of the
Hashing. Anyway, Rdb performed better than Ingres in that Benchmark.
Thanks for the responses.
Best Regards, Jeffrey
|