Mark,
-----Original Message-----
From: Mark Townsley [mailto:townsley@cisco.com]
Sent: Tuesday, July 28, 2009 10:44 AM
To: Templin, Fred L
Cc: IPv6 Operations
Subject: Re: Tunnel MTU
Thanks for keeping us on our toes Fred!
I'd like to clarify of course that some of the MTU issues we discussed
were not specific to tunneling, but to mismatched MTUs on a home LAN
vs.
WAN interface in general. The extra 20 bytes of a 6rd or 6to4
encapsulation isn't significant when trying to solve support of 9K
jumbo
frames and standard 1500 byte ethernet MTUs in the same network.
Two issues here. First, if the home link has a 9KB MTU
and the CPE advertises something like 1480, all nodes
on the home link will be stuck using a 1480 MTU even
for communications that never leave the link. At least
that's how it is in current IPv6 implementations.
Secondly, if the home network has multiple links (connected
by routers), the links that are deep inside of the home
network will not see the CPE router's MTU advertisement
and will continue using the native link size (e.g., 1500
bytes). Their 1500 packets will then of course be dropped
by the CPE router and an annoying ICMP PTB sent back to
trigger an even more annoying retransmission.
Tunneling does affect MTU, no doubt. I'll point out that with 6rd, the
deployment space of the tunnel is limited, making it dramatically
easier
to "tune" than with 6to4. Unlike SEAL, this is still essentially an
operator tweaked value in the current deployment model.
As I said in the meeting, the tunnel gives you a
configuration knob with 1280 at the "safe" extreme
and 1480 at the "living dangerous" extreme (unless
of course you know that *all* links in the ISP network
configure a larger MTU).
Another concern is what to
do about a misconfigured link somewhere in the ISP
network (e.g., it sets an MTU of 900 when it meant
to set 9KB). SEAL will ride over that without loss
and will even help identify the degenerate link so
it can be fixed. Without SEAL, setting a static MTU
even as small as 1280 causes undetected fragmentation
that can lead to severe problems especially when we
have anycast in the mix.
To go one step further, let's say we have home network
A and home network B with 9KB links behind CPE routers
running 6RD over an ISP network that is known to include
1500 links. With SEAL, nodes in home network A can send
9KB IPv6 packets to nodes in home network B *even though
the ISP network only supports 1500*. For that matter,
SEAL can also support IPv6 jumbograms (>64KB) over the
ISP network for those home users that truly want to push
the limits.