The authors certainly have a lot of experience in building RPC-based grid applications. However, the document in review 
seems to show only the very essence of their results, which makes it hard to follow. One of the questions I always had 
was: why is any existing RPC system (CORBA??) not doing the job you need to get done? I am absolutely convinced that the
 authors have a good answer, but it is not given in the document.

Another thing is that the current presentation takes a lot of my imagination for guessing how the API operations would 
be used. Unfortunately, the document does not point to a "use case study" from which the motivation for the taken API 
decisions could be motivated. Does the group
have such a document?

What kind of applications do you ave in mind? What kind of languages do you want ot support? (I'd guess that Java or C++
 could be much more elegant with exceptions etc.)

I would encourage the authors to provide some sample code (skeletons) that put their routines into context. E.g., what 
about some parameter sweep skeleton, or something more fancy??

about grpc_initialize: there is no explanation given about the purpose of the config file. about handle_default and 
handle_init: why have 2 functions?
Why have information about the repository in the application code at all?
Isn't the config file the place to put that?

text comments:

page 1 "status of this memo", needs to mention recommendations

page 2, "general workflow": I don't understand this paragraph