Cassio Neri wrote:
Hi Juergen and Björn,
sophisticated mathematical models -> costly comutations -> did you have
thought about grid computing, computing power on demand and highly scalable.
In short you can prepare a job containg the stuff needed to do your
calculations. Transfer it on the grid and run it. Data can be loaded from a
database or from some where else via some special and secure data channels
and can be stored back. It depends on your onw configruation how many CPU
and sub jbs in parallel are runnig ...
Ok it is a little bit off topic but i think interesting. Becasue you can
trigger such jobs from calc difectly and can feed the data into the sheets
for later use ... Or using webs wervices to get furher necessary data, stock
trades ...
The models are always probabilistic and, almost always, we use Monte
Carlo simulations to run them. This is, indeed, very good for
concurrency: In the heart of the algorithm there is a for loop where
each iteration is (or might be) completely independent from the
others. Normally those computations are done by an in-house developed
library. An add-in link the library to the spreasheet application
which works more as user interface. Only simpler computations (if any)
are done by the spreadsheet application itself.
There are other front-ends as well. In line with Juergen's suggestion,
for instance, we have XML parsers which load the data necessary to the
simulation from a database and generating XML reports with the
results.
There are, indeed, libraries exploiting the concurrency but,
unfortunately, I've no hands-on experience. More recently people
started to look at GPU computing as well. This is still a very new
topic with promising results. Once I came across an article on
Bloomberg's experience with GPUs computing. They were very pleased
with the scalability. Apparently, it's faster, cheaper and greener
than a farm of servers.
Talking about feeding data into the sheet, I have a question about the
type XVolatileResult (hope this is the right forum). Suppose, for
instance, that the Monte Carlo simulation takes 10 minutes. I was
wondering if we could use XVolatileResults to present in a cell the
progress of the simulation before it finishes. With Excel we can just
launch, wait and pray that something important is happening (some
libraries use a console window to monitor the progress). If so, then
can we stop the computation before it finishes?
yes, that is possible. You can feed the results that are already
available directly in the feed or you can return a percentage number.
The SDK contains an add-in demo that demonstrates a simple counter for
example.
Juergen
Regards,
Cassio.
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@openoffice.org
For additional commands, e-mail: dev-h...@openoffice.org
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@openoffice.org
For additional commands, e-mail: dev-h...@openoffice.org
