On the Cost of Supporting Mobility and Multihoming Mobility and Multihoming Vatche Ishakian, Ibrahim Matta, Joseph Akinwumi Computer Science Boston University I. Matta 1
Mobility = Dynamic Multihoming Mobility Dynamic Multihoming � Hosts / ASes became increasingly multihomed � Multihoming is a special case of mobility � Multihoming is a special case of mobility � RINA (Recursive InterNetwork Architecture) is a � RINA (Recursive InterNetwork Architecture) is a clean-slate design – http://csr.bu.edu/rina � RINA routing is based on node addresses � RINA routing is based on node addresses � Late binding of node address to point-of-attachment � Compare to LISP (early binding) and Mobile-IP � Compare to LISP (early binding) and Mobile IP � Average-case communication cost analysis � Simulation over Internet like topologies � Simulation over Internet-like topologies I. Matta
What’s wrong today? one big, flat, open net one big, flat, open net Web, email, ftp, … Applications Applications TCP, UDP, … TCP, UDP, … Transport Transport Transport Transport IP protocol IP t l Network Network Network Data Link Data Link DL DL DL DL Physical Physical PHY PHY www.cs.bu.edu www.cs.bu.edu 128 197 0 0 128.197.0.0 128 10 0 0 128.10.0.0 128.197.15.10 � There’s no building block Th ’ b ildi bl k � We named and addressed the wrong things (i.e. interfaces) � We exposed addresses to applications � We exposed addresses to applications
RINA offers better scoping Web, email, ftp, … Applications Applications Transport Transport Transport Transport TCP, UDP, … TCP UDP IPC Layer Network Network IP Network Data Link Data Link DL DL DL DL IPC Layer IPC Layer Physical Physical PHY PHY � The IPC Layer is the building block and can be composed � An IPC Layer has all what is needed to manage a “private” network, A IPC L h ll h t i d d t “ i t ” t k i.e. it integrates routing, transport and management � E2E (end-to-end principle) is not relevant � Each IPC Layer provides (transport) service / QoS over its scope � Each IPC Layer provides (transport) service / QoS over its scope � IPv6 is/was a waste of time! � We can have many layers without too many addresses per layer
RINA: Good Addressing – private mgmt Bob want to send message to “Bob” “Bob” � B A B IPC Layer I 2 I I 1 To: B IPC Layer � Destination application is identified by “name” � Each IPC Layer is privately managed � It assigns private node addresses to IPC processes � It internally maps app/service name to node address � It internally maps app/service name to node address 5
RINA: Good Addressing - late binding Bob want to send message to “Bob” A B IPC Layer To: B B, , are I 1 I 2 IPC processes I 1 I 2 on same on same IPC Layer B � I 2 machine � Addressing is relative: node address is name for lower IPC � Addressing is relative: node address is name for lower IPC Layer, and point-of-attachment (PoA) for higher IPC Layer � Late binding of node name to a PoA address � A machine subscribes to different IPC Layers 6
RINA: Good Routing source destination � Back to naming-addressing basics [Saltzer ’82] � Service name (location-independent) � node name (location-dependent) � ( p ) PoA address (path-dependent) � path � We clearly distinguish the last 2 mappings � Route : sequence of node names (addresses) � Route : sequence of node names (addresses) � Late binding : map next-hop’s node name to PoA at lower I. Matta 7 IPC level
Mobility is Inherent Mobility is Inherent MH CH � Mobile joins new IPC Layers and leaves old ones � Mobile joins new IPC Layers and leaves old ones � Local movement results in local routing updates 8
Mobility is Inherent Mobility is Inherent CH � Mobile joins new IPC Layers and leaves old ones � Mobile joins new IPC Layers and leaves old ones � Local movement results in local routing updates 9
Mobility is Inherent Mobility is Inherent CH � Mobile joins new IPC Layers and leaves old ones � Mobile joins new IPC Layers and leaves old ones � Local movement results in local routing updates 10
Compare to loc/id split (1) Compare to loc/id split (1) � Basis of solutions to the multihoming issue � Claim: the IP address semantics are overloaded as both C a t e add ess se a t cs a e o e oaded as bot location and identifier EID x -> EID y � LISP (Location ID Separation Protocol) ’06 EID x � EID y RLOC 1x � RLOC 2y EID x � EID y Mapping: EID y � RLOC 2y I. Matta 11
Compare to loc/id split (2) Compare to loc/id split (2) � Ingress Border Router maps ID to loc, which is the location of destination Egress BR � Problem: loc is path-dependent, does not name the ultimate destination EID x -> EID y RLOC 1x RLOC 2y EID x � EID y Mapping: EID y � RLOC 2y 12
LISP vs. RINA vs. … LISP vs. RINA vs. … � Total Cost per loc / interface change = Cost of Loc / Routing Update + Cost of Loc / Routing Update + ρ [ P cons *DeliveryCost + (1-P cons )*InconsistencyCost ] ρ [ y ( cons ) y ] cons ρ : 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 � 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 in ol es ro te propagation o er D/2 h h is tree height involves route propagation over D/2 h , h is tree height I. Matta 13
14 I. Matta LISP LISP
15 I. Matta LISP LISP
16 I. Matta RINA RINA
17 I. Matta RINA RINA
8 1 I. Matta RINA RINA
19 I. Matta MobileIP MobileIP
LISP vs. RINA vs. … LISP vs. RINA vs. … 8x8 Grid Topology RINA uses 5 IPC levels; on average, 3 levels get affected per move LISP RINA I. Matta 20
Simulation: Packet Delivery Ratio Simulation: Packet Delivery Ratio RINA � BRITE � BRITE generated 2- level topology � Average path length 14 hops � Random walk LISP mobility model � Download BRITE from www cs bu edu/brite www.cs.bu.edu/brite I. Matta 21
Simulation: Packet Delay Simulation: Packet Delay LISP LISP RINA I. Matta 22
Bottom Line: RINA is less costly Bottom Line: RINA is less costly � RINA inherently limits the scope of � RINA inherently limits the scope of location update & inconsistency � RINA uses “direct” routing to destination node � More work: prototyping p yp g I. Matta 23
RINA papers @ p p @ http://csr.bu.edu/rina Thank You Questions? Questions? I. Matta 24
Recommend
More recommend
