Comments on Message Exchange Pattern: Single Request-Response document

I have two comments on the MEP SRR document. 
 
There is discussion at the end of the document (in the section
"Transport Binding Framework Assumptions" onwards) concerning the
"environment", which is defined as "a container for those properties
that are not scoped on a per-message or per message exchange basis".
This could includes items such as session keys and certificates that are
needed for multiple  message exchanges and could be expensive to
acquire. Unfortunately none of this discussion has reached into the main
body of the MEP SRR document - and it should be there, as it will
substantially improve performance where multiple message exchanges
occur. 
 
The second point is that the processing model is very simplified - it
assumes the responding node will completely receive all the data in the
incoming message before even starting to process it. It is likely that
all efficient implementations will start processing each part of the
message as soon as it arrives. Some might not be able to do anything
until the end of the request has arrived, but some bindings can react
without having received the end of the incoming message. If the request
message is somehow incorrect, such bindings could send back a
pre-emptive error, thus preventing the sender having to transmit the
rest of the (faulty) request message. If the request msg is e.g. 1 MB
(e.g. with photo attachments) and the sender can be told quickly that
there is something wrong with it, then this prevents wasting bandwidth
and speeds up the sender in sending the subsequently correct message. 
 
Eamon O'Tuathail

Received on Tuesday, 4 December 2001 07:54:13 UTC