RE: [nsp] a question CRC error and resent

From: Kinczli Zoltán (Zoltan.Kinczli@Synergon.hu)
Date: Fri Apr 20 2001 - 02:48:30 EDT


  IMHO that's a retransmission of a L2 frame, not a TCP segment.
Eventually, a TCP fragment might get re-transmitted, or part of it,
depending the L2 frame size vs TCP segment size.

  Gert has right: the router will *never* retransmit a
packet(L3)/segment(L4).
Jim has right as well: the router might retransmit a frame(L2). But
those
retransmissions are very loosely coupled to each other :))

--Zoltan

-----Eredeti üzenet-----
Feladó: Jim Warner [mailto:warner@cats.UCSC.EDU]
Küldve: Friday, April 20, 2001 12:02 AM
Címzett: dsinn@microsoft.com; gert@greenie.muc.de; tatsuya@kivex.com
Másolatot kap: cisco-nsp@puck.nether.net
Tárgy: Re: [nsp] a question CRC error and resent

> The router will *never* retransmit a packet, except if the
> router is also the end station of the TCP connection
> ("telnet to router").

"Never" is a pretty strong word. If the underlying link protocol
calls for hop-by-hop acks, then packets are buffered until
ack'd and retransmitted as necessary. The list of link protocols
that do NOT do this is pretty long: ethernet, token ring, frame
relay, SMDS, HDLC encapsulated sync serial...even localtalk.

A counter example where retransmission is done hop-by-hop would
be X.25. Conventional analog modems (V.32, V.90) normally run
in error correction mode that causes "stuff" that arrives
damaged to be retransmitted. There may or may not be any
relation between the transmission units in your modem and
IP packets. But if you jam an analog modem into a router and
then run PPP over a dial-up connection, the router could fairly
claim to be doing hop-based retransmission over the phone line.

-jim warner, UCSC



This archive was generated by hypermail 2b29 : Sun Aug 04 2002 - 04:12:35 EDT