Transporting QinQ across a Layer 2 link locked at 1518 octets AND across a Layer 3 link

Mon Sep 13 13:00:02 UTC 2010

Only input I can give is on EOIP tunnels and MTU.

Scenario is we had 1 remote router running, Tunneled back to our network 
using EOIP so that the remote network could use our ip space. Don't 
remember how I figured it out, But I was using the MTU of 1458 (On the EOIP 
interface). Everything was great, But weird things started happening. 
Certain sites wouldn't load, MSN.com....etc it was strange. Jacking it back 
up to 1500 flat fixed it.

Nick Olsen
Network Operations
(877) 804-3001  x106

----------------------------------------

From: "Francois Menard" <francois at menards.ca>
Sent: Sunday, September 12, 2010 10:13 PM
To: "Francois Menard" <francois at menards.ca>
Subject: Re: Transporting QinQ across a Layer 2 link locked at 1518 octets 
AND across a Layer 3 link

Oops two typos - sunday evening casualties.

On 2010-09-12, at 10:06 PM, Francois Menard wrote:

> Folks,
> 
> Question #1:
> 
> Is it possible for me to put an MPLS router on both ends of a circuit 
leased from a transport service provider which does not support QinQ (i.e. 
packets of 1526 bytes), and which requires us to tag traffic onto a well 
specified set of VLANs (i.e. if we want two VLANs, the service provider 
tells us which two VLANs to use).
> 
> I was thinking of lowering the MTU size on my MPLS router such that I 
could do QinQ over VPLS, over MPLS, over Ethernet transport locked at @ 
1514 octets.
> 
> I would imagine that my effective IPv4 payload would be reduced to 
something like (not taking into account CRC removed in the Ethernet 
driver)
> 
> 1514 minus 8 bytes for MPLS label minus 4 bytes for VPLS control word 
minus 4 bytes for VLAN tag #1 minus 4 bytes for VLAN tag #2 = 
> 1514 -8 -4 -4 -4 -4 = 1490 octets
> 
> So if I set my MTU on my MPLS router at 1490 octets and send QinQ over 
VPLS over , wouldn't that allow for all of the above mentioned overhead to 
pile-up without exceeding the 1514 octets size allowed by my transport 
provider ?
> 

So if I set my MTU on my MPLS router at 1490 octets and send QinQ over VPLS 
over MPLS over Ethernet, wouldn't that allow for all of the above mentioned 
overhead to pile-up without exceeding the 1514 octets size allowed by my 
transport provider ?

> Question #2:
> 
> I have another link, which is restricted by the transport service 
provider, which is an MPLS-VPN service, and which does not support QinQ, 
nor supports layer 2 bridging.
> 
> An option available to me is to use an EoIP tunnel on a Mikrotik RouterOS 
router, which maps Ethernet over GRE over IP, causing some 28 octets of 
overhead. This is proprietary to Mikrotik.
> 
> So in this case, assuming that I want to do something as dangerous as:
> 
> QinQ over VPLS over MPLS over Ethernet over EoIP (over GRE, over IP)
> 
> And accordingly set my MTU to:
> 
> 1480 (from above) minus 28 octets (Ethernet over GRE over IP) = 1452 
octets 
> 

1490 (from above) minus 28 octets (Ethernet over GRE over IP) = 1462 
octets

> So if I set my MTU on my MPLS router at 1452 octets, wouldn't that allow 
for all of the above mentioned overhead to be successfully transported 
across an IP layer 3 circuit, effectively ending up with
> 

1462 octets

> QinQ over MPLS over Ethernet over IP ?
> 
> What are the consequences of setting the MTU as low as 1452 octets?  What 
applications end-up breaking ?
> 

1462 octets

> -=Francois=-
> 
> 
> 
> 