Transport protocols and bulk file transfer

[stanislav shalunov <>]

Participating:

        Larry Dunn
        Stanislav Shalunov
        Shawn McKee
        David Lapsley
        Injong Rhee
        Yunhong Gu
        Xinwei Hong

Not participating:

        Steven Senger -- meeting
        Simon Leinen
        Lisong Xu -- meeting

Review of previous action items

        Gu: Evaluating UDT copyright.  Currently considering LGPL.

        Stas: The actual license in the distribution tarball is not
        GPL as mentioned before, but a BSD-style license that's
        probably compatible with BSD.  LGPL is not compatible with BSD
        license.

        Larry: So, is the current license fine and there's no need to
        change it?

        Stas: The current license looks fine.  If it is changed, it
        should not be LGPL, but rather BSD license without advertising
        clause.

        Steven Senger (not present) is working on the API document.

        Gu: UDT provides sending rate, receiving rate, RTT, ACKs,
        inter-packet time, congestion window size.  User-defined
        control events for ACK packets, etc.  By default, UDT
        congestion control.

        Larry: Can UDT API be adopted as is?  Is UDT a candidate?  Is
        it the candidate?

        Stas: UDT is one of the code bases (and sets of ideas) that
        are being considered for being used in the new tool.

        Stas posted two new versions of the design space survey.
        Remaining work includes substantiating section 8, and, more
        importantly, writing conclusions.

>         * All: read the design space document and comment.

        Gu: The present document outlines choices, but makes no
        decisions.  Can we add decisions?

        Stas: That's what I meant by conclusions.

        Injong: Guy Almes is now at NSF, right?  What are the funding
        opportunities that are coming up?

        Stas: I don't know, but can ask Guy.

        Injong: NeTS might not be interested, since the tool is for a
        narrow scope of uses (high-performance).  Also, NeTS is very
        TCP-oriented and very competitive.

New participants

        Larry: I lead the advanced architecture group at Cisco.
        Interested in high-performance stuff.  Stas suggested that I
        join this group to help out in any way I can, and here I am.

Design space document comments

        Stas: Would prefer to make the decisions in the design
        document that can be made without extensive experimentation
        before the end of the year.

        Injong: Not sure about that.  Many of the questions have no
        clear answers.  (For example, rate-based vs window-based and
        loss-based vs delay-based.)  Better right and late than wrong
        and early.  Let us be realistic.

        Stas: The thing I thought we should try to do before the end
        of the year is to identify the questions we can answer (and
        answer them) and the questions that we can't yet answer.

        Larry: In many cases, experimentation and writing code are
        required to get an answer, but as Gu mentioned earlier, how do
        you write code without answers?  Perhaps tunable knobs.

        Injong: For some yes-or-no questions, the answer might be
        ``yes and no'' (perhaps configurable).

        Gu: Would like to see a requirements section in the design
        space draft.

        Injong: Requirements section would help understand the context
        for the decisions.

        Gu: Let's define environment and make decisions.

        Stas: There was a set of requirements in the beginning of the
        original document/proposal:
        http://www.internet2.edu/~shalunov/tmp/200404-bulk-transport-tool.pdf
        Is that the kind of requirements that we are talking about?

        Larry: Should the protocol be adaptive and run better in
        high-performance environment and no worse in standard
        environment?

NSF Networking Technology and Systems (NeTS) Program Solicitation

        Injong: Neutral, but would participate in the proposal if it
        were written.

        Shawn: Could help with putting the proposal together, but
        would not participate in it.

Action items

        Stas sends out call for decisions on design

        Stas talks to Guy Almes

        Stas asks others about NeTS

        Larry will send more comments to the list

        Steven S. keeps working on the API document

Next meeting

        Stas: If we hold the meeting earlier, we'd have more luck with having
        a permanent scribe join.  The candidate time slots are Monday 2PM and
        Friday noon (New York time).  Would this work for people on the call
        and is there a preference?

        Everyone is OK with moving the meeting to either of the new times.
        Shawn somewhat prefers Friday noon.  If we have people from Europe,
        noon might also be better.

        Stas: The meeting next week is canceled because it falls on
        the day after Thanksgiving.  Tentative new time for the
        meetings is noon on Friday (New York time).  Talk to you in
        two weeks.

        Call concluded at 15:58.

        [I apologize for the quality of these notes.  I'm not good at
        typing and listening or talking at the same time.  I probably
        got your words wrong, so send me corrections.  --Stas]

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

"I didn't attend the funeral, but I sent a nice letter saying that I
approved of it."                                -- Mark Twain