Transport protocols and bulk file transfer

[Scott Brim <>]

I won't make it to the call on Friday.  I'll think of you when I'm
sitting on the beach.  I have a few notes from the IETF, but no
follow-up actions yet:

  Kathie Nichols, "CoDel" queue management: This looks pretty good,
  and something we could use in our IP forwarding.  She wants to get
  her stuff into production.  She asks if we could talk to our
  vendors.

  In the endpoints, CAIA Delay Gradient TCP looks very interesting for
  overall throughput when we don't have intermediaries to help.
  However, Kathie thinks it has holes in it.  This should be watched,
  as the wizards argue it out.  The implementation they have is
  FreeBSD.  They were encouraged to do it for Linux.

  Google (Yuchung Cheng) suggests tightly coordinating TCP connections
  between the same source and destination - use a single congestion
  window, round trip time, loss rate, reordering, etc.  He says SCTP
  won't work because it does congestion management at the
  server/sender side (in a web environment) and he wants to do it at
  the client(?).  This should be watched, it may be a good idea.

  ECN (explicit congestion notification) would work with both but
  requires changes in both endpoints and forwarders.  It's been
  improved for more accuracy but it still works better in data centers
  than in the wide area.