> The main task of the BL is communications between instances of Piper. > Emerging tasks are resource management and security. > > Some issues that come to mind atm are: > 1. Resource localisation (a central registration server that keeps > track of Piper instances) > 2. Offered data streaming (push) > 3. Requested data streaming (pull) > > Other stuff we'll need, and that's outside the scope of kqml, are > 4. Node relocation (numero uno pain in the butt) > 5. Authentication (probably Akenti, but maybe Sesame) > 6. Resouce management (like ram, HD space, ea) Ok, let me rephrase that: You want the BL to manage both the process distribution aspects and the node coordination aspects. * 1, 2 and 3 are directory services (who does what and where) * 4 and 5 belongs to the node coordination topic * I'm still not clear what 6 is. Two more question: * are you sure you want 4 and how good do you want it? * are you sure you want to use a centralized approach for 1? The main comment I have is that this looks like we need to do some reading and look at distributed architectures, for the above subjects have been worked on already. I can't suggest a solution as I don't know what kind of information the directory srevices would have to publish and query, nor what 4 and 6 precisely are. > Ic. This is bad indead. darn.. really bad. Maybe that's why all the > code I could find was made 5-6 years ago. Depending on what you actually need, you may get away with something like RPC-over-something-else. Ever thought of SOAP maybe? Narval is using the python implementation by pythonware.com. Slow, but standard. -- Nicolas Chauvat http://www.logilab.com - "Mais où est donc Ornicar ?" - LOGILAB, Paris (France)