28
Some Thoughts on Loc/ID Separation DaeYoung Kim [email protected]

Some Thoughts on Loc/ID Separation

Embed Size (px)

Citation preview

Some Thoughts on Loc/ID Separation

DaeYoung Kim [email protected]

Questions

• Is LIS properly positioned?

• Any insights to a better network architecture?

IAB Routing & Addressing Workshop 2006

• Explosion of DFZ routing table

• << Incompetence of multi-homing

• << Semantic overloading of IP address

• >> Separation of Loc and ID

Semantic Overloading

Host

IP address

id

locator

Loc/ID Separation (LIS)

Host

Loc

ID

MigrationISP1

nw prefix subnet host i/faddr

Migration

*losing all sever connections at transition

renumbering*

ISP2

nw prefix subnet host i/faddr

Multi-homing

*Competing for more specific prefixes into DFZ

backup

majorISP1

ISP2

*You want PI addr

Loc doesn't help• Same fate as IP address

• Renumbering at migration

• Tend to be injected into DFZ

• Doesn't help mitigate table explosion

• LIS, not a solution to the problem

Fallacy of IP address

Host

IP addressPoA*

MAC addressL2

L3

*Naming the same leg twice

Fallacy of IP address

H

C

X

?

Different Locators

Host

LISP

ILNP

subnet

sitesubnet

subnet

HIP

What's up?

• Loc/ID separation?

• Recursive addressing?

• host addr - subnet addr - site addr

Norms…?

• Shach[’78]

• name(what), address(where), route(how)

• Saltzer[’78 & rfc1498]

• service, node, PoA, route

Links first

Links first

Links first

Links first

Nodes first

Nodes first

LISP EID• IP address? ... !

• Flat? ... !

• Static? ... !

• Globally unique? ... !

• no changes in host behaviour? ... !

if EID = IP address• Not flat

• hierarchically assigned PoA address

• changes as PoA changes

• Not globally unique

• need a new EID at migration to a new LISP site

• Semantically overloaded

Let EID == node ID• Pointing to the node(host, router; system)

• not to the interface

• Use ISIS for IGP, and

• In-site host/subnet mobility, a default

• no need for Mobile IP

ISIS

Hsubnet

subnet

subnet

R

Area Sys IDaddr

Area

EID global? Why not local?

• Name already global

• Server identified by a name

• name <-> [RLOC + (local) EID]

• Client given a temporary id anyway; DHCP

Recursive Addressing

tier 0 tier +1 tier +2tier -1

EID RLOC ??

Name

Any insights to a better network architecture?

Merci!