GSR12000-fragmentation on L2/L3VPN

[fukusho]

I'm a little bit confused about mtu fragmentation in mpls environment.

How does GSR12000 determine if the packet is L2 frame (for EoMPLS) or L3 frame(for RFC2547) at the egress mpls port?

For example,let's say the the port is configured as below;

fa0/0-->AToM(eompls)port (L2VPN)

fa1/0-->vpn-forwarding port (L3VPN)

Gig2/0-->mpls forwarding port

*IP MTU is set to 1500 for all ports.other mtu's are set to default.

If ip mtu with 1500 enters from fa0/0 and fa1/0, how does GSR understand that the packet is L2 or L3 frame at Gig2/0?

I was wondering if GSR is looking inside the frame for IP header or MAC header to determine this at the egress whether to fragment it or not when it exceeds the egress mtu. (Are these different in other cisco boxes?)

[Harold Ritter]

If fragmentation is required, the content after the bottom of the label stack is examined and the following actions are taken depending.

If the content is l3vpn and that the DF bit is not set, the packet is fragmented and forwarded toward the egress PE.

If the content is l3vpn and DF bit is set, an ICMP "fragmentation needed but DF bit set" message is forwarded to the IP source via the egress PE (using the original LSP) , which will in turn forward it to the originator of the packets needing fragmentation. The reason to forward via the egress PE is that the core probably doesn't have an entry in its routing table for the destination of the ICMP message.

If the content is l2vpn, no fragmentation takes place.

Hope this helps,

[fukusho]

Thank you.

Just few more to clear things out.

Ok, understood that the bottom of label stack is examined. I was wondering how does it judge that it's IP header or MAC header with just 1's and 0's? Does it read the hole 20bytes?

I was under the impression that examining too much information may cause performance problems so that routers check the least significant bit(if there are any...I'd like to know).

I understand that mpls network should basically be fragment free, but was just curious..

[Harold Ritter]

The way IOS differentiates between l2vpn and l3vpn is implementation specific. Lets's just stick with "it looks into the information after the label stack" for the purpose of this discussion. From a performance standpoint, this is really not an issue any more with asic based forwarding.

You are absolutely correct about this. You should really keep your MPLS network as fragmentation free as possible.

Hope this helps,