1. I. Matta 1 On the Cost of Supporting Multihoming and Mobility Ibrahim Matta Computer Science Boston University Joint work with Vatche Ishakian, Joseph Akinwumi, John Day

2. I. Matta Mobility = Dynamic Multihoming  Hosts / ASes became increasingly multihomed  Multihoming is a special case of mobility  RINA (Recursive InterNetwork Architecture) is a clean-slate design – http://csr.bu.edu/rina  RINA routing is based on node addresses m Late binding of node address to point-of-attachment  Compare to LISP (early binding) and Mobile-IP  Average-case communication cost analysis  Simulation over Internet-like topologies

3. What’s wrong today? Network Transport Data Link Physical Applications Network Transport Data Link Physical Applications Network DL PHY Web, email, ftp, …  We exposed addresses to applications  We named and addressed the wrong things www.cs.bu.edu 128.197.15.10 128.197.15.1 128.10.127.25 128.10.0.0 128.197.0.0

4. RINA offers better scoping Network Transport Data Link Physical Applications Network Transport Data Link Physical Applications Network DL PHY TCP, UDP, … IP Web, email, ftp, … IPC  E2E (end-to-end principle) is not relevant m Each IPC layer provides service / QoS over its scope  IPv6 is/was a waste of time! m We don’t need too many addresses within an IPC layer

5. 5 RINA: Good Addressing  Destination application is identified by “name”  App name mapped to node name (address)  Node addresses are private within IPC layer m Need a global namespace, but not address space m Destination application process is assigned a port number dynamically BA I1I1 I2I2 want to send message to “Bob” IPC Layer To: B “Bob”  B Bob IPC Layer

6. 6 RINA: Good Addressing  Late binding of node name to a PoA address  PoA address is “name” at the lower IPC level  Node subscribes to different IPC layers BA I1I1 I2I2 want to send message to “Bob” BI2BI2 To: B Bob IPC Layer B,, are IPC processes on same machine I1I1 I2I2

7. I. Matta 7 RINA: Good Routing  Back to naming-addressing basics [Saltzer ’82] m Service name (location-independent)  node name (location-dependent)  PoA address (path-dependent)  path  We clearly distinguish the last 2 mappings  Route: sequence of node names (addresses)  Map next-hop’s node name to PoA at lower IPC level sourcedestination

8. 8 Mobility is Inherent  Mobile joins new IPC layers and leaves old ones  Local movement results in local routing updates CHMH

9. 9 Mobility is Inherent  Mobile joins new IPC layers and leaves old ones  Local movement results in local routing updates CH

10. 10 Mobility is Inherent  Mobile joins new IPC layers and leaves old ones  Local movement results in local routing updates CH

11. I. Matta Compare to loc/id split (1)  Basis of any solution to the multihoming issue  Claim: the IP address semantics are overloaded as both location and identifier  LISP (Location ID Separation Protocol) ‘06 EID x  EID y EID x -> EID y EID x  EID y RLOC 1x  RLOC 2y Mapping: EID y  RLOC 2y

12. Compare to loc/id split (2)  Ingress Border Router maps ID to loc, which is the location of destination BR  Problem: loc is path-dependent, does not name the ultimate destination EID x -> EID y EID x  EID y RLOC 1x RLOC 2y Mapping: EID y  RLOC 2y

13. I. Matta LISP vs. RINA vs. …  Total Cost per loc / interface change = Cost of Loc / Routing Update +   [P cons *DeliveryCost + (1-P cons )*InconsistencyCost]  expected packets per loc change P cons: probability of no loc change since last pkt delivery  RINA’s routing modeled over a binary tree of IPC layers: update at top level involves route propagation over the whole network diameter D; update at leaf involves route propagation over D/2 h, h is tree height

14. I. Matta LISP

15. I. Matta LISP

16. I. Matta RINA

17. I. Matta RINA

18. I. Matta RINA

19. I. Matta MobileIP

20. I. Matta LISP vs. RINA vs. … RINA 8x8 Grid Topology RINA uses 5 IPC levels; on average, 3 levels get affected per move LISP

21. Simulation: Packet Delivery Ratio  BRITE generated 2- level topology  Average path length 14 hops  Random walk mobility model  Download BRITE from www.cs.bu.edu/brite I. Matta 21 RINA LISP

22. Simulation: Packet Delay I. Matta 22 LISP RINA

23. I. Matta Bottom Line: RINA is less costly  RINA inherently limits the scope of location update & inconsistency  RINA uses “direct” routing to destination node  More work: prototyping

24. I. Matta RINA papers @ http://csr.bu.edu/rina Thank You Questions? http://csr.bu.edu/rina