LISP Data Plane Suggestion.

The data plane in LISP will add (encap) 36 bytes before the ipv4 packet. This is a traditional GRE tunnel mechanism. There are 3 portions , Outer header, UDP, LISP header. For doing this, router need more resource to accomplish it. For JNPR, I think the Tunnel PIC will be a mandatory to put more header in original packet. But how about 2GB per second traffic into the Tunnel PIC or into RP ? There should be some way to do the this like MPLS label. I would like to put my words on LISP Data Plane.

After control plane, the ITR know how the IP address of the ETR. The ITR will encap new header before original packet. But How about change a new way ?

New Header for LISP

[Dst-ipv6] + [Src-ipv6] + [exp bit] + [S bit]

The Dst-ipv6 is combination format like
"2010:dst-ipv4 of destination locator: dst-ipv4 of original packet"

The Src-ipv6 is combination format like
"2010:src-ipv4 of source locator: src-ipv4 of original packet"

This format also can be aggregate like ipv4 does. Because each PI can only adv one ipv6 CIDR which will covery all. For example ASN 100 adv 2010:210:210:16::/64, Like ipv4, no one will care the dst-ipv4 of original packet. The router only need to know where the next hop of this CIDR is point to.

In this format, all internet Tier1, Tier2 router must running ipv6 routing which is smaller then today's BGP table.

PHP

Like MPLS, this new format also can save performance issue as PHP do in MPLS. So this new format can put the information to upstream which indicate your are the last hop router to reach source locatior. You need to strip the header. This will save routing look up at source locator.

The new format new to be check with AISC team, they might calculation on each bit for the header. It could be a good way may be not. It's my 2cet.

LISP overview

What is LISP ? It's Locator/ID Separation Protocol from Cisco system. This a new mechanism which will change the size of routing table inside router. First, the core router will not care EID anymore. By doing this, ingress and egress router acts the key point in LISP. Howevr the MR, MS proxy also import.

The encap could be a big performance impact of the service. The first probe packet in the flow might be missing on the way to dest ip. And the multi-vendor support still not get answer from others(Juniper, 3com, etc)

From the LISP, we can see that CSCO put a new way on next generation routing table , but JNPR still make a big machine put everything bigger and bigger(memory, CPU, Disk).

Visit the website for detail.

http://lisp4.cisco.com/index.html