Re: summary and follow-up: transparent tcp improvements

From: Eric Travis ([email protected])
Date: Tue Apr 16 2002 - 20:32:58 EDT

  • Next message: Brian Levine: "Networked Group Communication 2002 (Call for Papers)"

    Karl,

    >The FOLLOW-UP question is:
    >Is there experience with tcp spoofing, increasing the throughput
    >from the sender over the wired network towards the gateway, and
    >removing slow-start (and possibly modifying congestion avoidance)
    >(by the gateway) towards the mobile terminal behind the (long delay
    >
    >and fully controlled) wireless link ?
    >tcp sender -- (network) -- gateway -- (wireless link) -- mobile receiver
    >
    >(no modifications on sender/receiver, low bandwidth and long dely on
    >the wireless link, many small tcp connections as typically in http /
    >www-traffic)
    >

    What you are looking for seems to be an entity that be
    can dropped in at the interface between your wireless
    links and their interconnection to the/a relatively
    "uncontrolled" internet (intranet); Via deployment of
    this magic box you'd like:

       a. To garner as much bandwidth over the tethered
          network path as if your mobile client were
          co-located with this entity

       b. To maximize the data throughput over the tetherless
          path > In The Direction Toward The Mobile Device <

       c. To leave as unmodified the protocol stacks (and
          any other configuration specific parameters) on
          both the mobile device and the random data sources
          from which the mobile wishes to extract data/content

    Since you state that the tetherless portion of your
    network path is "fully controlled", the magic box you
    seem to desire should (keyed to the above):

       a. Act as a transparent proxy for the mobile device
          (either at the application layer or a the transport
          layer); All associated distastefulness and pitfalls
          acknowledged[*] by you and related to your users/subscribers
          (right?)

       b. Pace the data (in the form of proper TCP segments)
          toward the mobile device at a rate prescribed by
          your "fully controlled" wireless link's configuration
          parameters.

       c. The above pacing should be *reliable*, making use
          of (and properly interpreting) the well defined TCP
          semantics - as the mobile device is going to be
          utilizing an unmodified TCP.

       d. Bonus: Retransmit any segments lost over the
          tethered network (for whatever reason) without
          requiring them to be retransmitted over the
          tetherless path (improving both power utilization
          and responsiveness)

    If I have gotten the above correct, then to answer your question:

       Yes, we do have experience doing precisely this;

       Yes, it works very well (but also introduces all
       the risks associated with the violence of shoving
       something in the middle of a TCP connection)

    Since the mailing lists are not places for advertising
    or plugs (shameless of otherwise), I've previously sent
    you appropriate pointers/contact information for a
    TCP Tranquility (SCPS-TP) gateway implementation.

    Remember as you go down this path:

      since you are accreting transport state at the
      interface of the wireless/wired regions, you are
      going to have to have some mechanism(s) (outside
      the scope of your transport protocol/spoofing
      techniques) for handing off this state (these
      transport connections) as your mobile roams.
     
    Eric
    ([email protected])

    [*] unfortunately for those who choose to live
        an honest life, there are no free lunches.



    This archive was generated by hypermail 2b29 : Tue Apr 16 2002 - 20:38:56 EDT