Re: [v6ops] Quick measurement: MTUs on the general Internet

[Jeroen Massar <[e-mail]>]

On 2014-12-18 15:57, Iljitsch van Beijnum wrote:
> I left the tcpdumps running for almost a week and did a more thorough analysis of the results, see:
> 
> Maximum packet sizes on the internet 
> http://www.bgpexpert.com/article.php?article=151

"my server received 41753 incoming TCP SYN"
..
"Turns out, most of the IPv6 traffic on my server is from bots that check "

Sample size too small to represent the Internet which also does not
consists out of bots....

...
"1280 and 1480 are probably IPv6-in-IPv4 tunnels and 1428 AYIYA tunnels."

If you have the addresses I can tell you exactly what type of tunnels
they are if coming from SixXS space[1], though likely it does not matter
too much what type it is.

As you are looking at MSS, most clients are configured that the tunnel
has a MTU of 1280 while the local network announces a MTU of 1500 and
thus a MSS of 1500-overhead.

Hence, you will see an MSS for 1500 and then get a PTB from the PoP the
tunnel terminates on for the first reply packet.

MSS clamping on tunnel PoPs is not being done as that would hide
problems with any other protocol than TCP and thus would make debugging
a real hell "TCP works, but hey, UDP does not WTF!" etc...

Greets,
 Jeroen

[1] https://www.sixxs.net/pops/prefixes/