Basis of Multiple-Participant decision making

[smidl]

Multiple Participand decision-making (MPDM) was the main reason why to start this project.

I propose to keep any communication and related stuff out of BDM library. All stuff related to mpdm will be located in folder applications/mpdm where classes such as "participant" or perhaps "agent" should reside.

Rather than writing the whole class hierarchy by us I proppse to look for a free version of multiagent systems - written preferably in C++. I am not sure about Java... would it be OK?

• make survey of multi-agent systems
• choose one system and copy (?) it into mpdm (?)
• build offsprings of their agents that would use element of the BDM library for communication and for internal routines

[vbarta]

I'd prefer not to build a Java application on top of a C++ library - I've done it before, and while it's certainly possible, it requires a lot of glue to cover the language incompatibilities (types, exceptions, threads), with lots of space for very interesting bugs... Porting Java to C++ is OK in general, but I'm afraid multiple participants will want to communicate over a network, which is implemented in standard libraries in Java (multiple standard libraries - network communication in Java may have stabilized by now, but that doesn't mean all multi-agent systems will use the latest ad greatest way to communicate) and 3rd-party libraries in C++, and they have almost nothing in common... So when you're doing the survey, please try to find something in C++ - if there really isn't anything good enough, well, it is possible to use Java...

[smidl]

I came across an interesting page of open source projects from NASA featuring at least two interesting projects:

• http://people.nas.nasa.gov/~bgreen/growler/ (C++)
• http://www.nas.nasa.gov/Research/Software/Open-Source/CODE/overview.html (Java)

I am also in favor of C++, however, I would like to go as mainstream as possible...

[smidl]

After a short look at the NASA thing i am not really happy with them :(

Another option appears to be Mobile-C which is written in C/C++ but most likely in Ch which is a sort of interpreted C.

• http://www.mobilec.org/
• http://www.softintegration.com/

It looks very powerfu but also very scary (interpreted C?)...

[smidl]

As agreed, we will not try to fill an existing system, but rather create a trivial implementation of our own dumb-agent application.

Issues:

• data exchange format: most likeliy string in libconfig format,
  • memory management and exception handling - this needs to be resolved in BDM library
• communication mechanism: perhaps ACE?
• how to manage the agents? On a single machine, on multiple...

Perhaps, we could use JADE as inspiration, I believe that it is the most used environment.

[vbarta]

Upon reflection, I don't really recommend ACE ( http://www.cse.wustl.edu/~schmidt/ACE-overview.html ), but Boost seems to have Boost.Asio ( http://www.boost.org/doc/libs/1_39_0/doc/html/boost_asio.html ) as a modern ACE replacement... That would also take care of memory management, using the same library.

[smidl]

On a second thought, the amount of work needed for implementation of such a beast is too high to end up with only a proof-of-concept solution...

Hence, I did a bit more search. This time not on agent-systems but "distributed control". I think this is much more fruitful than before. There are various implementations such as TACO, SMI++, but the most promising looks to be TANGO: http://www.tango-controls.org/

Advantages:

• actively maintained
• developed in Europe by research institutions, -> potentially active collaboration on European level
• written as opensource (binaries for debian)

Disadvantges:

• appears to be too complex to our current needs

However, I think it is worth trying before reinventing the wheel...