Transport protocols and bulk file transfer

[stanislav shalunov <>]

Steve,

Here's my set of nits, ahead of Monday's deadline for the first action
item of last teleconference:

0. Title, so that the document can be handed out in Arlington.

1. X_SOCKET probably also contains data structures necessary to
   identify the underlying UDP socket and any buffering data
   structures.  Or is that elsewhere?

2. I'm not quite sure how to implement x_listen() and x_accept().  It
   would seem to me that, with a UDP socket underneath, we'd need to
   get rid of x_listen() and to only allow one connection per port,
   unless we want a user-level demux process looking at the source
   address and handing the incoming packet to the right place.  The
   extra process would mean doubling the number of context switches
   and more complex software architecture that requires some kind of a
   daemon (presumably forked off in x_listen()?).  The straightforward
   path would seem to be to get rid of x_listen() and only have
   x_accept() work once on a given socket (and return that socket
   itself or maybe an integer success indicator).  Would that be a
   burden from the point of view of application programming?

3. Regardless of resolution of (2), should x_accept() only take one
   argument?  In Berkeley sockets API, the addr argument for accept()
   is a result argument; we could tuck that away inside X_SOCKET
   structure.

4. In x_setsockopt(), how does one set reliable datagrams vs
   unreliable datagrams behavior?  The QTTL option is about partial
   reliability.  Unreliable service could be set with 0; for reliable
   service, a large value could be used.  Perhaps the default should
   be mentioned here as well.  Should the value be u_int32_t (making
   it unsigned and fixing the width)?  Or perhaps at least unsigned
   int?  Also, should the value be in microseconds for better
   resolution?  2^32 microseconds is more than an hour, which is
   plenty.

5. In x_setsockopt() and x_getsockopt(), do we need the
   optname/optval/optlen system?  Could we just give it a union?

6. In x_socket(), should x_sockstatus() be x_sockerror()?

-- 
Stanislav Shalunov              http://www.internet2.edu/~shalunov/

Clothes make the man.  Naked people have little or no influence on
society.                                             -- Mark Twain