3-2_routing-ad hoc 09

Document Sample
3-2_routing-ad hoc 09 Powered By Docstoc
					Wireless Ad hoc networks
 – Routing
Proposed ad hoc Routing Approaches
• Conventional wired-type schemes (global
  routing, proactive):
   – Distance Vector; Link State
• Proactive ad hoc routing:
   – OLSR, TBRPF
• On- Demand, reactive routing:
   – DSR (Source routing), MSR, BSR
   – AODV (Backward learning)
Wireless multihop routing challenges
 • mobility
 • need to scale to large numbers (100’s to
   1000's)
 • need to support multimedia applications
   (QoS)

 • unreliable radio channel (fading, external
   interference, mobility, etc)
 • limited bandwidth
 • limited power
Conventional wired routing limitations
• Distance Vector (eg, Bellman-Ford, BGP):
   – Tables grow linearly with # nodes
   – routing control O/H linearly increasing with
     network size
   – convergence problems (count to infinity);
     potential loops (mobility?)
• Link State (eg, OSPF):
   – link update flooding O/H caused by network size
     and frequent topology changes
CONVENTIONAL ROUTING DOES NOT SCALE TO
  SIZE AND MOBILITY

           DV     LS
Intra-AS   RIP    OSPF
Inter-AS   BGP
      Proactive ad hoc schemes
         – OLSR and TBRPF
• Link State explodes because of Link State update
  overhead
• Question: how can we reduce the O/H?
• Answer: Link State with “Topology reduction”
   – (1) if the network is “dense”, use fewer
     forwarding nodes
   – (2) if the network is dense, advertise only a
     subset of the links
• Two leading IETF Link State schemes enhance
  scalability in large scale networks:
   – OLSR : Optimal Link State Routing
   – TBRPF: Topology Broadcast Reverse Path
     Routing
          LSR (Link State Routing)
• In LSR protocol a lot
  of control msg
  unnecessary
  duplicated




                          24 retransmissions to diffuse a
                          message up to 3 hops


                           Retransmission node
    OLSR (Optimal Link State Routing)

• In OLSR only a
  subset of neighbors
  (MPR-Multipoint
  Relay Selectors)
  retransmit control
  messages:
   – Reduce size of
     control message;
   – Minimize flooding   11 retransmission to diffuse a
                         message up to 3 hops


                             Retransmission node
              OLSR Overview
• RFC 3626, October 2003
• In LSR protocol a lot of control messages
  unnecessarily duplicated
• In OLSR only a subset of neighbors (MPR-Multipoint
  Relay Selectors) retransmit control messages
   – Reduce flooding overhead
   – Adapted for dense network
• OLSR retains all the advantages of LSR:
   – stable;
   – Does not depend upon any central entity;
   – Tolerates loss of control messages;
   – Supports nodes mobility
  On-Demand Routing Protocols
• Routes are established “on demand” as
  requested by the source
• Only the active routes are maintained by
  each node
• Channel/Memory overhead is minimized

• Two leading methods for route discovery:
  source routing and backward learning
  (similar to LAN interconnection routing)
      Existing On-Demand Protocols
•   Dynamic Source Routing (DSR) -- CMU
•   Multipath Source Routing (MSR) – TJU
•   Backup Source Routing (BSR) – UofO+TJU
•   Ad-hoc On-demand Distance Vector (AODV)
•   Associativity-Based Routing (ABR)
•   Temporarily Ordered Routing Algorithm (TORA)
•   Zone Routing Protocol (ZRP)
•   Location assisted routing (LAR, DREAM)
•   Signal Stability Based Adaptive Routing (SSA)
•   On Demand Multicast Routing Protocol
    (ODMRP) – UCLA
  Dynamic Source Routing (DSR)
• RFC 4728 – February 2007
• Forwarding: source route driven instead of
  hop-by-hop route table driven
   – Mobility ?
• No periodic routing update message is sent
• The first path discovered is selected as the
  route
• Two main phases
   – Route Discovery
   – Route Maintenance
         DSR - Route Discovery
• To establish a route, the source floods a Route
  Request message with a unique request ID
• The Route Request packet “picks up” the node
  ID numbers
• Route Reply message containing path
  information is sent back to the source either by
   – the destination, or
   – intermediate nodes that have a route to the
     destination
• Each node maintains a Route Cache which
  records routes it has learned and overheard
  over time
     DSR - Route Maintenance
• Route maintenance performed only while
  route is in use
• Monitors the validity of existing routes by
  passively listening to acknowledgments of
  data packets transmitted to neighboring
  nodes
• When problem detected, send Route Error
  packet to original sender to perform new
  route discovery
   MSR - Multipath Source Routing
• Direct Descendant of DSR
• On-demand + Source Routing + Multipath
• Probing-based adaptive load balancing among
  multiple paths
• Motivation of MSR
  – Efficiently using the network resource
  – Alleviate the oscillation in adaptive single
    path routing
  – Fast re-routing
  – Reducing computing & storage requirement
  – Exploiting computing power of host instead
    of link capacity
Distributing Traffic among Multiple
               Paths
• Quantities: A heuristic equation
• Probing-based adaptive control
   – Decoupling between transport layer and
     network layer: SRPing
   – Cost effective
• Scheduling: Packet Weighted Round Robin
• TCP out-of-order (re-sequencing) problem
          Distributing Traffic among
                Multiple Paths
• Heuristic equation
   – Rationale: Autonomous system, homogeneous
     assumption, bandwidth-delay product constant
                           j  
                     min   d max  ,U   R
                j
            W   i
                           d j  
                               i      
where ,
    j
 d i is the delay of route with index i,
   j
 d m axis the maximum delay of all the routes to the
  same destination,
    R is a factor to control the switching frequency
  between routes.
    U is a bound value to insure that should not to be
  too large.
             MSR Summary
• Reduce network congestion
• Improve throughput, delay, mobility, fault
  tolerance (CBR & FTP)
• Acceptable routing overhead?
   – Little more than that of DSR
   – Route discovery
   – Route maintenance
      • Probing (unicast) add little O/H
• Good candidate for QoS support
   – QoS-MSR, reliable-MSR
• Acceptable packet out-of-order level ?
   Backup Source Routing (BSR)
• Establish and maintain backup routes that
  can be utilized after the primary path breaks
• Define a new routing metric - route
  reliability, and use it to provide the basis for
  the backup path selection
• Reduce the frequency of route discovery
  flooding, which is a major overhead in on-
  demand protocols
• Can improve the performance significantly
  in more challenging situations of high
  mobility
      Simulation Methodology
• ns – Wireless extensions by CMU
• Adopt methods used in [Broch98, Johnson99]
• Two major files:
   – Movement pattern file
   – Communication pattern file
• 50 mobile hosts placed randomly within a
  1500m×300m area
• 20 connections
• Different traffic types: CBR & FTP
• Two set of simulations: Max speed 20m/s &
  1m/s
     Performance Evaluation
• MSR vs. DSR vs. BSR
• Performance Metrics
  – Packet delivery ratio
  – Data throughput
  – End-to-end delay
  – Packet drop probability
  – Queue size
Simulation Results with UDP Traffic
      -- Packet delivery ratio for 20 sources




                                                8
                                   Simulation Results – CBR
• End-to-end throughput

                             200
                                                                            DSR
                                                                            MSR
Throughput, packets/second




                             150




                             100




                              50




                               0


                                   1   2   3   4   5   6   7   8    9   10 11 12 13 14 15 16 17 18 19 20
                                                                   Connection No.
Simulation Results with UDP Traffic
   -- Average end-to-end delay for 20 sources




                                                11
                          Simulation Results - CBR
• Packets dropped at each node

                                                                                      DSR
             15
                                                                                      MSR
# of drops




             10




              5




              0
                  1   3   5   7   9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
                                                       Node No.
Previous Work on Using Multiple Paths
 • Alternate use (primary and backup)
   – It works OK for CBR traffic (BSR, Bypass -
     DSR, Node Disjoint M-path AODV, etc)
   – TCP does not get much benefit. Backup path
     is used only after timeout; not efficient in
     mobility/errors.?
 • Concurrent use (ie, packet scattering)
   – MSR
   – TCP does well in a static, error free net with
     long paths (up to 50% improvement)
   – With mobility & errors, TCP suffers out-of-
     order problems because of RTT difference
     on the two paths
“TCP Performance on multiple paths
   in ad hoc nets..” Liaw et al ICC 2004




Static net, no errors, opt W: max improvement 50%;
typical improvement between 8% and 18%
Multiple Path TCP with Packet Replicas
• TCP data packet duplication on multiple paths
   – May introduce less O/H than repeated end
     to end retransmissions
• Improve end-to-end route robustness when
  single route is not stable:
   – Replicate packet on multiple paths
   – Combat random, non correlated link losses
   – Combat path breakage
                           Variable Loss Rate [ 0.05; 0.1; 0.15; 0.2]
Total Throughput(bits/s)




                                             Mobility(m/s)
                              Original TCP                   Multipath TCP

				
DOCUMENT INFO
Shared By:
Categories:
Stats:
views:4
posted:7/13/2011
language:English
pages:28