I am more or less a newbie concerning OLAP. Anyway, I am writing a
thesis about this topic and found some - in my eyes - contradictions
about write-back possibilities of MOLAP in contrast to ROLAP.

Some sources say it is only possible with MOLAP (especially in
connection with what-if scenarios), some say, you should only use in
with ROLAP. Are there different kinds of write-back scenarios? Could
anybody give an explanation of this?

Michael Huhn
Germany

It depends on what product you're using, rather than being a MOLAP vs
ROLAP issue.

For example, in Analysis Services, both MOLAP and ROLAP cubes can handle
write-back, though the transactions themselves are held in a ROLAP
partition. But many ROLAP products (such as MicroStrategy) are intended
mainly for sales analysis, and don't themselves support write-back. Of
course, there are other ways of updating relational tables, it's just
that the ROLAP tool itself doesn't handle it. Other ROLAP-based
financial products do support write-back because their target
applications require it.

So your conclusion should be that write-back capabilities are a function
of the application, and have little to with how the data is stored on
disk. If your app needs it, both MOLAP and ROLAP can support
write-back, but if the product isn't indended for use in such apps, it
probably won't offer it, regardless of whether it's MOLAP or ROLAP.

<mhuhn.de (AT) gmail (DOT) com> wrote

While ROLAP or MOLAP doesn't inheritly limit the possibility of
write-back, Michael should be aware of the performance implications of
different architectures. The ability to perform the write-back
operation, typically implies that the user would like to see the
results of the changed data reflected in the other derived figures
(aggregated and calculated values). Memory based archtectures
generally offer much better support (i.e. performance) for these type
of interactive sessions.

Dave Menninger
Applix