ietf
[Top] [All Lists]

Re: [mpls] [lisp] draft-ietf-mpls-in-udp was RE: gre-in-udp draft (was: RE: [tsvwg] Milestones changed for tsvwg WG)

2014-01-14 10:14:17

One of the reasons that IPv6 (rfc2460) dropped the header checksum
that was in IPv4 is that with link layers all doing FCS it served no
purpose.
 For the same reason TCP and UDP checksums server no purpose.

That's not what I understood, it is needed for endpoint delivery
verification. But in this case, TCP & UDP checksums incorporate the IP
address info in the pseudo header, and this was thought sufficient.

The link layer checksum fails and the packet is dropped.  Since this
is usually not at the last hop (often a local Ethernet) but instead in
a WAN link, the packet never arrives at the destination for anything
to count IP header or TCP or UDP checksum errors.

Again, that presumes that everything above the link works flawlessly,
which isn't always the case.

Also all of the following handle both data integrity and congestion
avoidance the same way:

  UDP over IP over Ethernet
  UDP over IP over MPLS over Ethernet
  UDP over IP over MPLS over UDP over IP over Ethernet
  s/Ethernet/{POS,GFP,etc}/ for all of the above
  s/^UDP over IP/PW/ for all of the above

In all of the above cases data integrity is handled by the link layer.
In all of the above cases congestion avoidance is handled by the
application (or not at all).  Whether MPLS is carried directly over a
link layer or over UDP/IP over a link layer makes no difference.

RFC 6936 section 3.1 says more about what happens when packets happen to
be mis-delivered to the wrong host or socket.

Curtis


Gorry

In message
<290E20B455C66743BE178C5C84F1240847E63346BF(_at_)EXMB01CMS(_dot_)surrey(_dot_)ac(_dot_)uk>
l(_dot_)wood(_at_)surrey(_dot_)ac(_dot_)uk writes:

Really, you'd want to expose the pseudoheader check at endhosts; a
well-instrumented Linux box could tell you a lot
about checksum failures.

But in this case, a router would be decapping UDP/MPLS tunnels as an
endpoint, so could report on checksum failures -
if the checksum wasn't zero.

Lloyd Wood
http://about.me/lloydwood
________________________________________
From: Dino Farinacci [farinacci(_at_)gmail(_dot_)com]
Sent: 12 January 2014 21:37
To: Wood L  Dr (Electronic Eng)
Cc: <mark(_dot_)tinka(_at_)seacom(_dot_)mu>; <mpls(_at_)ietf(_dot_)org>; 
gorry(_at_)erg(_dot_)abdn(_dot_)ac(_dot_)uk;
lisp(_at_)ietf(_dot_)org; david(_dot_)black(_at_)emc(_dot_)com; 
randy(_at_)psg(_dot_)com; tsvwg(_at_)ietf(_dot_)org;
jnc(_at_)mit(_dot_)edu; ietf(_at_)ietf(_dot_)org
Subject: Re: [lisp] [mpls] draft-ietf-mpls-in-udp was RE: gre-in-udp
draft (was: RE: [tsvwg] Milestones changed for tsvwg WG)

Do any routers count TCP/UDP checksum failures, much less
expose the count via SNMP?

Typically they do but only for packets destined to them. Much like hosts
would check the header checksum.

Dino
_______________________________________________
mpls mailing list
mpls(_at_)ietf(_dot_)org
https://www.ietf.org/mailman/listinfo/mpls



<Prev in Thread] Current Thread [Next in Thread>