Re: Relationals vs. Objects Databases I

Kevin Dick wrote:

> Andrew McClure wrote:
> >
> > I have been following this thread with some interest, and being relatively
> > new to Object Databases, I won't fly into the current debate, however -
> > I have one simple question. If one was to take a variety of running
> > RDBMS applications, and port them to an ODBMS on similar hardware -
> > Is there a rule of thumb that would predicate the performance
> > difference, would it favour ODBMS, and if so, by how much.
> >
> > Further to this, if one did the same for RDBMS systems enhanced with,
> > and using Object extensions, would these results differ dramatically?
>
> The performance improvement would depend a lot on the data model and access
> pattern of the application. Application data models with lots of relationships
> between elements and a traversal access pattern would experience a significant
> performance increase from a switch to an ODBMS. From benchmark results and
> anecdotal evidence, this improvement ranges from 1 to 3 orders of magnitude
> decrease in response time. What is "lots of relationships"? Hard to say, but
> if you have an E-R diagram where most of the arcs cross more than one other arc,
> that's a pretty good indication.
>

This is a bunch of BS. Where did you get this figures. They might be close to the truthonly if you compare VERY poor relational data model with VERY good object model.
I've already contributed to this thread at the begining of this discussion and I will repeat -
It depends. If your queries are pure pointer navigations then yes, you will get in general certain performance gain (something like 3-5 times). But if you need pattern
search or interval comparisons then forget it. RDBMS will outperform ODBMS by a huge
margin. I won't give you any figures because they heavily depend on vendor, your implementation of the problem, platform and database engine tuning, data distribution, etc.
For example: you have just three object classes - customers, transactions and inventory.
Number of instances are 20000, 50000 and 5000000 respectively which I would say relatively modest. You need to select customers by geographic regions who ordered within certain period of time product of certain categories with total sale greater then
certain amount. I would like to see how ODBMS will perform in this case AND, which is
important but almost always ommitted by ODBMS propaganda, how much time and efforts will it take to create such a query. With RDBMS it is a single query - 6 -7 lines in old
plain SQL or few keystrokes and mouse clicks with many of readily available query and
report tools.

> It's much harder to say for ORDBMS because their support for object features
> differs significantly. In most cases, the underlying storage is still tables so
> the improvement for relationship traversal should be modest, perhaps a factor of
> 2 decrease in response time at best. However, if the issue is complex
> datatypes, the results will be much better. In most cases, ORDBMS extend the
> datatype capabilities quite a bit. Storing and manipulating ADTs and multimedia
> types could yield an order of magnitude or more improvement.
>
> Kevin
Received on Thu Feb 05 1998 - 00:00:00 CST