[c-nsp] System MTU on trunks for Q in Q

[Dan Armstrong]

The reason I don't want to raise it too high - is if we're selling TLS 
services to a customer, (ie a VLAN provisioned on 2 ports on different 
switches, carried across our core/trunks) - I don't want them being able 
to send any packet larger than 1500 byes. 

A bit bigger wouldn't be a problem, but if I set it to, say 9000, and 
all of a sudden we have some jackass with a storage head that could be 
firing 9000 byte packet across our backbone... not cool, I believe it 
would cause havoc with "small packet" applications like VoIP, even with 
QoS, the bit-time to send a 9000 byte packet out an interface is 
significant.

I've also never been too clear on the interaction between the system mtu 
command, and the system mtu jumbo command.  I've always just made them 
match...





Peter Rathlev wrote:
> On Wed, 2008-03-26 at 19:01 -0400, Dan Armstrong wrote:
>   
>> I've been bashing my head against the wall all day for a definitive 
>> answer on this:
>>
>> On a Cisco switch that supports QinQ (3550, 3750, ME3400, 3560 etc)
>>
>>
>> What is the _minimum_ value I need to set the system MTU to, to do 
>> QinQ?  1504?  1522?  1526?  1546?
>>
>> I can't seem to find one concise answer...
>>     
>
> I'm not entirely sure what the "system mtu" specifies, i.e. if it's
> interface MTUs (typically excluding data link headers) or what.
>
> IPv4 packet = 1500 bytes. Ethernet header = 14 bytes. 802.1q header = 4
> bytes. Another one = 4 bytes more. So for "simple" QinQ of "regular"
> IPv4 traffic it would be max 1522 bytes of data per packets.
>
> Any special reason not to just raise it to the maximum of 2000 bytes?
>
> Regards,
> Peter
>
>
>