Page 1 of 39

Parallel RNG and distributed MC

Posted: January 25th, 2012, 12:45 pm
by Cuchulainn
Discussion of TRNG and Mersenne Twister (parallel) here I think it is important that we do not invent the wheel.

Parallel RNG and distributed MC

Posted: January 25th, 2012, 5:42 pm
by Alan
You REALLY need a new (non-moving) icon -- that one is a huge visual distraction. To add something on-topic, I am interested in anybody's attempt to use cloud services for parallelism, esp. multiple GPU instances.I can google on my own, so much more interested in actual experiences than being pointed at links. p.s. Thanks! -- so much easier on the brain now.

Parallel RNG and distributed MC

Posted: January 25th, 2012, 6:25 pm
by Alan
mj has a nice paper that is pretty explicit about how to implement something on a local machine with a specific Nvidia card.What I would like is to avoid buying the hardware and somehow just get the thing to run remotely at amazon cloud or someplace. So, really, all I need is a single remote machine with a single GPU card -- and step-by-step instructions!

Parallel RNG and distributed MC

Posted: January 25th, 2012, 7:57 pm
by Cuchulainn
For MC201 I would like to plug in the 2 bespoke rng models in C++, first under Windows and then a port to gcc. Even now I can plug it into MC102 to see how it works.Issues like installation (autotool) and test examples would help as well. Any ideas on the delivery date(s)?

Parallel RNG and distributed MC

Posted: January 25th, 2012, 8:27 pm
by Alan
QuoteOriginally posted by: outrunMaybe a simple way forward -independent of activities here- is to use *Matlab* with "jacket" (the parallel GPU extension for Matlab) an run that at amazon.Link Do you have a link to MJ's paper?Sure, Graphical asian optionsNot a Matlab user and I am just interested in cheap, easy, experimentation at this stage...

Parallel RNG and distributed MC

Posted: January 25th, 2012, 8:31 pm
by Cuchulainn
QuoteAnd than we can probably align the SDE, payoff but from your C# 102, keep that compatible?That could be very intersted as we could make a C++/CLI interface (from C++) to your code.BTW will it be a boost library? It entails replacing this code by the new hybrid interface. All we need is to know what your C++ interface will be.

Parallel RNG and distributed MC

Posted: January 26th, 2012, 11:11 am
by DevonFangs
A stupid contribution from my side: I noticed that a much appreciated "order zero" is just distributed computed via MPI. People like it because it's very simple: you have to do 10**5 simulation and you have 4 machines, just make 25k on each machine and then aggregate.I think it'd be useful if an interface for this were provided. People could build small grids using the machines they have in their group for simple tasks.