Print

[r1ske]

FYI we are still looking into it as a configuration option inside the broker. There are just other things that require our attention at the moment.

Any word on this?  The ability to set a higher MTU could help with transfer rates and overall performance.  Count me among the interested.  Thanks in advance!

[malfeasant]

personally, i think this is a non-issue, mtu should be automatically negotiated to the highest possible, meaning that of the link with the lowest.  I suspect brad's problem may be due to a certain howto, http://wiki.debian.org/DebianIPv6, which lists:

auto 6in4
iface 6in4 inet6 v4tunnel
  address [Your IPv6 Endpoint]
  netmask [Prefix Length]
  endpoint [PoP IPv4 Endpoint]
  ttl 64
  up ip link set mtu 1280 dev 6in4
  up ip route add default via [PoP IPv6 Endpoint] dev 6in4

... or something similar
I'm curious whether that's just a typo, or if there is a reason for it...
but what do i know, i'm just a n00b 

[brad]

personally, i think this is a non-issue, mtu should be automatically negotiated to the highest possible, meaning that of the link with the lowest.  I suspect brad's problem may be due to a certain howto, http://wiki.debian.org/DebianIPv6, which lists:

auto 6in4
iface 6in4 inet6 v4tunnel
  address [Your IPv6 Endpoint]
  netmask [Prefix Length]
  endpoint [PoP IPv4 Endpoint]
  ttl 64
  up ip link set mtu 1280 dev 6in4
  up ip route add default via [PoP IPv6 Endpoint] dev 6in4

... or something similar
I'm curious whether that's just a typo, or if there is a reason for it...
but what do i know, i'm just a n00b 

Should be except that is not how tunnels work, and I don't even run Debian never mind a Linux-based OS.

[r1ske]

I tried setting the MTU of the tunnel interface on my router to 1480, but whenever I do a PMTU discovery over it the tunnel server drops it back down to 1280.  MTUs are not automatically negotiated with 6in4 tunneling and must be set manually on both sides.  Incidentally, 1280 is the smallest allowed IPv6 packet according to RFC 2460, but a smaller MTU means fragmentation is more likely and can hinder performance.

[snarked]

Does HE support "Internet2" size MTUs in its network?  Please note that I'm NOT limiting my question to those using a tunnel server (since I also have NATIVE IPv6 with HE as the peering carrier).  I noted that here in Los Angeles, HE peers with cenic.net (which uses Internet 2 MTUs).  Or is 1,500 bytes a limit in HE's transit network?

Transit peering link:
   10gigabitethernet2-2.core1.lax1.he.net (2001:470:1:1b::1)
   lax-hpr--he-peer.cenic.net (2001:468:e00:801::1)
(extracted from a traceroute6).

[kcochran]

1500 isn't a transit network limitation, we use jumbo frames wherever possible.  Most of the links in LA, for example are 9212.

[snarked]

Thanks for the info.  Now I have to figure out how to do it via an ethernet interface (server to router).

[snarked]

Just found out from my co-lo facility that they can't get me onto a gigabit port within their internal network so I can reach HE's feed into there via a jumbo frame.  :-(  However, I still have a dedicated 100baseT IPv6-only ethernet cable/port connection.  Their Cisco router should handle it, but it's the intervening router that's the problem.  (However, when they upgrade, I'll be ready to go.)

Any harm leaving my MTU at 9216?  OK, so I'll get one ICMP packet back (from the internal network) and switch the MTU to 1500, but why not just leave it?

[sgucukoglu]

FYI we are still looking into it as a configuration option inside the broker. There are just other things that require our attention at the moment.

Any word on this?  The ability to set a higher MTU could help with transfer rates and overall performance.  Count me among the interested.  Thanks in advance!

<AOL>Me too!</AOL>

Any word?  Are you still on it?

I have a troublesome configuration: an Apple Time Capsule (or Airport) loses packets with IP payloads >1496 bytes.  That's 1496 +8 PPPoE = 1504 over-large Ethernet packet payload.  Sadly, BT, my ISP, insists on shoving large frames down peoples' throats, operating at 1500-byte MTU and relying on either jumbo frames or MSS clamping to make the connection useful.  So until I find hardware that supports baby jumbos >1504 payload, I'd need HE to let me tune the tunnel MTU to 1476.

But of course, you already know that. 

Cheers,
Sabahattin

PS: In theory, this should all be self-regulating, since HE can respond to IPv4 ICMP "Packet too big and DF set MTU=x" messages.  Not in this case, however.