Re: [tcpm] WGLC on draft-ietf-tcpm-early-rexmt-01

[Mark Allman <[e-mail]>]

> I don't like the part about tracking the seqnum edges.
> 
> Why would someone send 400 byte packets when the MSS is 1400+?

Something interactive like ssh.  Why do we give apps the chance to turn
off Nagle?  It's a big, ugly world with a vast range of behavior.

> How does this interact with the Nagle algorithm?

As sketched in the draft ...

  + You can use a byte-based variant of ER which assumes MSS-sized
    packets.  As discussed in the draft this *can* have precision
    problems in terms of determining the number of packets outstanding
    in the network and therefore when ER should and should not be used.
    In particular, if Nagle is turned off and so short packets are being
    sent then ER can get confused (see two examples in section 2.1.

  + Given that there can be a lack of precision we note that a TCP could
    in fact track the three packets on the right side of the window
    (four sequence numbers) to fully grok how many packets are in the
    network [*] and therefore to more accurately trigger ER.  This in
    particular helps when Nagle is in use.  I am not clear what you are
    getting at by an 'interaction'.  I don't think ER interacts with
    Nagle, but the packet-based variant does cope with Nagle.

    [*] Obviously tracking only the right side of the window doesn't
        give you a fine-grained notion of how many packets are in the
        network in total.  It tells you that either (a) there are at
        least four (and so we're out of the ER regime) or (b) there are
        less than four outstanding packets and how many there actually
        are.  So, it works for ER's purpose.

allman