On 13 May 2014 16:05, Mike Belopuhov <m...@belopuhov.com> wrote: > On 13 May 2014 15:45, Claudio Jeker <cje...@diehard.n-r-g.com> wrote: >> With KAME the MTU size of the loopback interface became strange and is >> actually dependend on the architecture. I see no point in all this >> just go back to the way it was long long long ago and just use 32k as the >> MTU. AFAIK all of this was only done to test large IPv6 packets but why >> MHLEN and MLEN were added is still beyond my imagination. >> > > In fact this comes from this commit in NetBSD originally: > http://cvsweb.netbsd.org/bsdweb.cgi/src/sys/net/if_loop.c#rev1.20 > > Comment says "Add MHLEN + MLEN extra space to LOMTU for IP and > transport headers." This doesn't make me understand why this > is needed however. So unless this clues you in, I'm OK with > setting it to 32k everywhere.
And purely for your amusement: FreeBSD have decreased MTU from 64k to 16k in '95: http://svnweb.freebsd.org/base/head/sys/net/if_loop.c?r1=6875&r2=6876 Perhaps (and I'm somewhat speculating here) because of TCP performance problems, like so: http://marc.info/?l=netbsd-current-users&m=102684186303106&w=2 (fixed diff is in the next mail though)