RE: Random seed (RANDOMIZE)

From: Chris Theis <Christian.Theis_at_cern.ch>
Date: Sat, 2 Jul 2011 06:31:51 +0000

Alberto,

I daresay that multi-core chips are not the latest fashion in computer
science as they've been around on the consumer's market now for >5
years. In addition I fully admit that I have observed their development
for quite some time before starting to parallelize some of my code.

I appreciate your penchant for illustrative comparisons from the
automobile industry (I very well remember the nice tire pressure
comparison). You're absolutely right - the FIAT 500 was a nice and
robust car. However, there is a new model now (which is by the way the
new CERN car) and it comes with air-condition and a plug for MP3
players. Given the choice I tend to say that people would choose this
one over the old model.

> FLUKA has just not been designed for fancy things as asynchronous SPMD, and I don't
> recommend that we spend precious human resources on it.

As a developer you have without a doubt in-depth knowledge of the code
structure and if you say that it is incompatible with modern
parallelization concepts then I fully agree with you that it's useless
and a waste of resources to re-write major parts from scratch. So my
question is fully answered for which I'd like to thank you.

Yet there is one more thing which I would like to comment on:

> This is not the case of M.C. for particle transport, where each particlehistory can have a
> different number of steps and each interaction a different number of
output channels (th > same reason why particle transport cannot be done
with Quasi-Monte Carlo: a pity,
> because it would be much more efficient).

Having worked on and with Quasi-Monte Carlo for quite some time I would
say that it is actually possible to do particle transport with it. There
are some limits and it cannot be used exclusively but there is a lot of
research going on in this direction as well as successful applications
and implementations to be found in hybrid designs. It usually requires
significant mathematical groundwork and the level of complexity added in
the implementation is obviously non negligible for those hybrid
solutions. In cases with a huge but fixed number of dimensions, like in
the proto-typical applications for financial mathematics, these methods
are much much simpler to incorporate. Yet, in some cases for transport
calculations the effort still pays off and for example in domains like
global illumination, for which you also have to solve transport
equations, it has been state-of-the art since a couple of years to use
MC-QMC hybrids.

Chris
Received on Sat Jul 02 2011 - 13:24:22 CEST

This archive was generated by hypermail 2.2.0 : Sat Jul 02 2011 - 13:24:22 CEST