SORT BY:

LIST ORDER
THREAD
AUTHOR
SUBJECT


SEARCH

IPS HOME


    [Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

    Re: TCP limitations (was Re: ISCSI: Urgent Flag requirement violates TCP.)



    
    
    Venkat,
    
    I might be mistaken but it the formula on which Vern's examples are built
    is based on the assumption
    that SACK or DSACK are not used.
    
    If SACK is used and the window is large enough to accomodate the RTT then
    your line of reasoning fails.
    
    
    
    Julo
    
    "Venkat Rangan" <venkat@rhapsodynetworks.com> on 21/11/2000 18:39:55
    
    Please respond to "Venkat Rangan" <venkat@rhapsodynetworks.com>
    
    To:   "Venkat Rangan" <venkat@rhapsodynetworks.com>, "IP Storage Working
          Group" <ips@ece.cmu.edu>
    cc:
    Subject:  Re: TCP limitations (was Re: ISCSI: Urgent Flag requirement
          violates TCP.)
    
    
    
    
    Julo,
    
    > I fail to see how your drop assumptions change the need for framing.
    > If we don't have framing we have to design the stack/adapter to either:
    >
    >   store all the data until the TCP recovers
    >   drop all the data until TCP recovers
    
    The amount of "anonymous" buffers you need to hold in the event of a
    sequence hole (and in the absence of Urgent Pointer), is also limited
    by TCP throughput achievable. Following Vern's example, if the amount
    of time for the receiving end to send an ACK for the hole and for the
    sending
    end to retransmit the missing segment is closely related to RTT, it would
    be
    about 100msec. Even if one were to assume that it is 200msec, given a BW of
    about 2MB/sec, you'll have to hold only about 400KB, which is not very
    significant.
    
    If one were to decrease RTT so we get higher BW (say 100us), you get a
    BW of 2GB/sec, but then for such an RTT, you need to hold in anonymous
    buffers, again
    only 400KB. This is again assuming that stacks at either end only add
    another RTT
    in processing time, for a total of 200us. The relationship of anonymous
    buffer
    size is really to 1/sqrt(p), and not the BW or the RTT per-se.
    
    Venkat Rangan
    Rhapsody Networks Inc.
    www.rhapsodynetworks.com
    
    
    
    
    
    


Home

Last updated: Tue Sep 04 01:06:18 2001
6315 messages in chronological order