Re: Burstiness WIth increased max cwnd.

From: mukul goyal ([email protected])
Date: Thu Feb 24 2000 - 21:44:27 EST

  • Next message: Zachary Amsden: "Re: Burstiness WIth increased max cwnd."

    No, I was not referring to burstiness after dropped segment.

    TCP may potentially send a large number of packets in rapid succession
    followed by a relatively large period of silence. This will happen if
    ACKs arrive in rapid succession. I am interested in studying this
    phenomenon and how frequent is this with both default and scaled window
    sizes.

    I was wondering if previous work exists on this topic.

    Also, I will be interested in knowing what are the burst control
    techniques currently being employed by various implementations.

    Thanks,
    Mukul

    On Thu, 24 Feb 2000, Kacheong Poon wrote:

    > > With Window scaling in TCP, the back-to-back packets a TCP flow sends
    > > can be very high. I was wondering if there is some study evaluating the
    > > increase in burstiness of TCP traffic with larger cwnd. Or, in general,
    > > are there some papers talking about how bursty the traffic is as seen by a
    > > router?
    >
    > Do you mean the bursty traffic after there is a dropped segment? If there is
    > no dropped segment (including ACKs), TCP is as bursty as it is without window
    > scaling, sending 3 segments for every ACK (assuming delayed ACK for every 2
    > segments). I believe many implementations now have some simple form of
    > bursty control. And with SACK and various forms of NewReno, I think TCP is
    > not as busrty as it used to be when there is no window scaling.
    >
    > K. Poon.
    > [email protected]
    >
    >



    This archive was generated by hypermail 2b29 : Thu Feb 24 2000 - 22:19:31 EST