Dear Friends and Dynare developers,

last week, when visiting Bank of Finland, I met with two representative from the firm Techila (http://www.techila.fi), Reiner Wehkamp, managing director, et Teppo Tammisto, R&D director.

They sell a middleware to facilitate parallel computing, mostly embarassingly parallelizable tasks, on an heteregenous infrastructure or on the cloud.

They provide an API for Matlab among several other languages.

This API is different from Matlab's parallel toolbox and specific macros need to be specified in the *.m files.

Their solution is much less expensive than Matlab Distributed Computing Server.

Their previous contacts with Bank of Finland made them interested in using Dynare on their middleware. They modified master_parallel.m as well as numgrad2.m and hessian.m.

They are interested in a collaboration with the Dynare project.

Technically, it wouldn't be very difficult. It could be as simple as they maintain a directory under ./contrib with the few *.m files that they need to modify and we check for a "techila" option when setting the path in dynare_config.m

Financially, it is a bit more complicated but they seem to be willing to support somehow the Dynare project if using Dynare with Techila helps them penetrate the central bank market.

The next step for them is to work on a proof of concept with Bank of Finland.

As it is the first time, that the Dynare project has been approached by a commercial firm for possible collaboration, I wanted to inform you of this development and am curious for your comments.

As an aside, during the demo, I have been impressed by the facility to deploy applications on the cloud (for example Amazon cloud) and the very low cost of mobilizing very large parallel infrastructures. I expect this type of resources to become increasingly popular within the academic community at least for those institutions that don't have access to a supercomputer or a large cluster. We should think on how to make it easy to use Dynare/Octave on such clouds (Matlab licencing makes it probably difficult and/or expensive). We could also use it for testing parallel features.

The CEF pre-conference workshop on GPU computing in Vancouver will use Amazon cloud with the participants. I will let you know how this works out.

All the best,

Michel