Take the 2-minute tour ×
Server Fault is a question and answer site for professional system and network administrators. It's 100% free, no registration required.

When configuring an IPv6 LAN it may be desirable to use router advertisements to announce a link MTU of either the standard 1500 bytes or something larger to enable usage of jumbo frames, perhaps 9000. This will allow communication between hosts on the LAN to use the largest frames possible. If the default route to the IPv6 internet must traverse a type of 6in4 tunnel, the MTU is usually 1480 or 1472 if PPPoE is in use.

Communication between LAN hosts should be uninterrupted, but the normal chain of events on traffic along the default route will behave differently. During the handshake for most connections, the first packet containing a significant amount of data will almost always be greater than the MTU of the default route, prompting the router to drop the packet and send an ICMPv6 packet too big message (type 2). I presume that most operating systems cache the results of path MTU discovery by destination address, and so this interaction will occur with nearly all default-route connections. This exchange should take only tens of milliseconds at most so I don't expect it to cause significant performance problems.

The question is this: Is this type of configuration considered best practice? Is it preferred to use a link MTU that is equal (or less than) the MTU of the default route path? Is there any documentation for this?

                    LAN          WAN

hostA -----\                     v4 WAN Link / 6in4 tunnel
            \  MTU 9000            MTU 1500     MTU 1480
             |---------- router ------------------------  -  -  -
            /                    IPv4 Internet, 6in4 endpoint -->
hostB -----/

          <<< router adv
                  prefix
                   RDNSS
                MTU 9000
                     etc

Example TCP connection:

TCP(SYN, 94 bytes)-------------------------------------------->
<----------------------------------------TCP(SYN/ACK, 86 bytes)
TCP(ACK/PSH, 1635 bytes)---X
<--------ICMP(too big, MTU=1480)
TCP(ACK/PSH, 1480 bytes)-------------------------------------->
...
share|improve this question

1 Answer 1

I think such a setup makes perfect sense. There will be many Packet-Too-Big (PTB) messages between the clients and the default gateway, but when you use jumbo MTU on your LAN then you probably do that because there is lots of traffic on the LAN anyway that can benefit from that. A few extra packets won't even be noticeable.

What I do see is that some consumer CPEs send an MTU size of 1480 (or 1472 etc) in the router advertisement. The impact on the LAN traffic won't be that much (max like 1.9%) and it will prevent PTB messages between the default gateway and the client at least, although they might still come from links further on the Internet that are even smaller. It depends on where your priorities are I guess.

I personally would always optimise the network and not worry about PTB messages at all.

share|improve this answer

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.