Robust Wireless Multicast using Network Coding

Document Sample
Robust Wireless Multicast using Network Coding Powered By Docstoc
					Robust Wireless Multicast using
       Network Coding


         Dawn Project Review,
           UCSC Sept 12, 06
              Mario Gerla
     Computer Science Dept, UCLA
gerla@cs.ucla.edu; www.cs.ucla.edu/NRL
Background – Network Coding
   Traditional multicast: store and forward




                                              2
Background – Network Coding
  Network Coding:store-mix-forward




                                     3
     Network Coding : wireless net
                      Store-mix-forward
              a   a   a                    a a,b b

      a                        a    a                      b


      a                        a    a                      b
                                            a         b
     a                         a   a,b
                                   a       a+b     a+b     b,a
                                                           b
           optimal routing               network coding
          energy per bit = 5           energy per bit = 4.5

 Wu et al. (2003); Wu, Chou, Kung (2004)
 Lun, Médard, Ho, Koetter (2004)

                                                                 4
      Random Network Coding

                                         Sender

        x          y    z                  Every packet p carries e = [e1
                                           e2 e3] encoding vector prefix
               A                           indicating how it is constructed
                                           (e.g., coded packet p = ∑eixi
αx + βy + γz                               where xi is original packet)

                                buffer




                                           Random
                                         combination

      Destination      Intermediate nodes randomly
                       mix incoming packets to
                                                                      5
                       generate outgoing packets
Robust NC Multicast
 Most studies have evaluated NC M-
  cast in static networks; no errors
 In tactical nets one must consider:
   Random errors; External
    interference/jamming
   Motion; path breakage
 Target application:
   Multicast (buffered) streaming
     Some loss tolerance
     Some delay tolerance (store & playback at
      destination) - non interactive

                                                  7
Network Coding in static wireless nets
 For cost efficiency
      Médard et al. “Min-cost operation over coded Networks.”
       IEEE T-IT
      Fragouli et al. “A network coding approach to energy
       efficient broadcasting…”, INFOCOM ’06
      Wu et al. “Minimum-energy multicast in mobile ad hoc
       networks using network coding.” IEEE TComm.


 For reliability
      Médard et al. “On coding for reliable communication over
       packet networks.”


 Others…
      Ephremides et al. “Joint scheduling and wireless network
       coding.” In Proc. NETCOD 2005.


                                                                  9
        NC vs Conventional M-cast
               comparison
 Conventional Multicast: ODMRP
   Mesh “fabric”; Redundant paths
   Robust to motion and to errors




                                     10
NC-Multicast evaluation
 Simulation study
   Scenarios with errors and motion
   Reported in IEEE Wireless Communication
    Magazine Oct. 2006 issue
 Performance bounds
   Static grid - “corridor” model
   Uniform, random errors
   Idealized MAC protocol (time slotting; non
    interfering sets of hyperarcs)
   Linear programming optimal solutions
   Manually computed optimal solutions
   Reported in MILCOM 2006
                                                 11
Simulation experiments
 Settings
   QualNet
    100 nodes on 1500 x 1500 m2
   5 Kbytes/sec traffic (512B packet) - light load
   Single source; multiple destinations
   Random Waypoint Mobility
   20 receivers
 Metrics
   Good packet ratios: num. of data packets received
     within deadline (1sec) vs. total num. of data packets
     generated
   Normalized packet O/H: total no. of packets
     generated vs no. of data packet received
   Delay: packet delivery time


                                                         12
ODMRP vs NC: Reliability

                     1.01
                       1
                     0.99
 Good Packet Ratio




                     0.98
  Delivery Ratio




                     0.97
                     0.96
                     0.95       CodeCast-8-dp0
                                CodeCast-8-dp10
                     0.94       CodeCast-4-dp0
                                UDP-dp0
                     0.93       UDP-dp10

                     0.92
                            0      10             20       30   40
                                      Max Node Speed (m/sec)
                                                                     13
ODMRP vs NC: Efficiency
                          3.5

                           3
 Normalized Packet OH |




                          2.5

                           2

                          1.5
                                                         CodeCast-8-dp0
                           1                             CodeCast-8-dp10
                                                         CodeCast-4-dp0
                          0.5                            UDP-dp0
                                                         UDP-dp10

                           0
                                0   10            20              30       40
                                         Max Node Speed (m/sec)                 14
ODMRP vs NC: Delay

                                 0.6
Average End-to-End Delay (sec)




                                 0.5
                                                           CodeCast-8-dp0
                                 0.4                       CodeCast-8-dp10
                                                           CodeCast-4-dp0
                                                           UDP-dp0
                                 0.3                       UDP-dp10


                                 0.2

                                 0.1

                                  0
                                       0   10        20             30       40

                                           Max Node Speed (m/sec)                 15
                    ODMRP vs. NC: Highway scenario
                 1.01                                                                            0.8

                   1
                                                                                                 0.7
                 0.99
                                                                                                 0.6




                                                                          Normalized Packet OH
                 0.98                                   NC-dp0
Delivery Ratio




                                                        NC-dp10                                                                       NC-dp0
                 0.97                                                                            0.5
                                                        ODMRP-dp0                                                                     NC-dp10
                 0.96                                   ODMRP-dp10                               0.4                                  ODMRP-dp0

                 0.95                                                                                                                 ODMRP-dp10
                                                                                                 0.3
                 0.94
                                                                                                 0.2
                 0.93
                 0.92                                                                            0.1
                 0.91
                                                                                                  0
                        0     10          20             30          40                                0   10          20            30             40
                                   Node Speed (m/sec)                                                           Node Speed (m/sec)




                            Randomly moving 200 nodes on 10kmx50m field.
                              All nodes are receivers.

                                                                                                                                               16
Robustness of NC approach




  Robust to random errors   Robust to mobility



                                                 20
Throughput Bounds
 Max NC-MCAST throughput in wireless networks?
      Previous simulation results based on light load. As load
       is increased, congestion leads to performance collapse
 Our approach: evaluate max throughput analytically
  for a simple grid structure, the “corridor”:




                                                              21
Linear Programming approach
 To calculate and compare maximum throughputs with
  and without NC, we use LP formulation

      Maximum multicast throughput LP models exist for
       wired networks
      We developed LP models for maximum throughput in
       unreliable wireless networks based on:
        LP model developed for min-cost problems in
          unreliable wired network by Muriel et al.
        wireless medium contention constraints
    Also, we solve with LP for max throughput of
     conventional multicast (single tree and tree packing)
    LP solutions matched with “manual” solutions




                                                          22
Related Work – Throughput Bound
 Previous works show the gap between NC
  and S/F for wired networks with no loss
  (e.g. log(n))

 For wireless networks
   Ephremides et al. “Joint scheduling and wireless
    network coding.” In Proc. NETCOD 2005.
   Wu et al. “Network planning in wireless ad hoc
    networks: a cross-layer.” IEEE JSAC 2005.
  => Both show throughput gain of NC
  calculated using link scheduling heuristics

                                                   23
Linear Programming Formulation
          maximize f


                   Wireless medium contention
                   constraints




                  Wireless flow conservation
                  constraints



                                               24
Maximum Multicast Throughput
Comparison: NC vs Conventional
CORRIDOR MODEL

    Sender                               0.7




                 End-to-end Throughput
                                         0.6




                   (Link Capacity=1)
                                         0.5
                                         0.4

                                         0.3

                                         0.2       Network Coding
   Receivers
                                         0.1       Multicast with Tree Packing
                                                   Multicast with Single Tree
                                          0
                                               0             0.1                 0.2
                                                   Link Error Probability



                                                                                   25
Network Coding: Link schedule achieving throughput of 2/3

    A                 B                          C                  D
                              A          B                               C           D

                                                        BA



    (1)         (2)                (3)           (4)          (5)             (6)
    E                 F                          G              H
                              E          F                               G           H

          D C                                           F E

A                         B                  C                       D
                                  A+B                                        C+D
    (7)         (8)                (9)           (10)         (11)            (12)
                                                                               26
Multicast with multiple embedded trees (no NC): Link schedule
achieves 2/5 throughput

A                   B


          A                        B

                         A                    B
                                                         A
    (1)       (2)            (3)       (4)         (5)

C                   D


          C                        D

                         C                    D
B                                                     C
 (6)          (7)            (8)        (9)         (10)
                                                                27
An “optimal” Single Tree multicast schedule that achieves 1/3


 A                                     B

              A                                     B

                            A                                    B


 (1)          (2)           (3)        (4)          (5)         (6)




                                                                 28
Future Work in Network Coding
 Implement NC - Mcast congestion
  control and ETE recovery above UDP
   If loss used as feedback, key problem is
    discrimination between random error and
    congestion
 TCP over Network Coded unicast
 Network Coding solutions for
  intermittent connectivity
 Models that include mobility

                                          29
   Vehicular Sensor Networks -
  Epidemic Dissemination Models
 Car-Car or Car-Infostation communications using DSRC
   DSRC: Dedicated Short Range Communication 802.11p IEEE
    Task group and derived from 802.11a




                      Roadside base station




                    Vehicle-to-roadside                 Inter-vehicle
                     communications                   communications



                      VSN-enabled vehicle
                       Sensors            Systems



                      Video   Chem.   Storage Proc.




                                                                        30
Vehicular Sensor Applications
 Environment
   Traffic congestion monitoring
   Urban pollution monitoring
 Civic and Homeland security
   Forensic accident or crime site
    investigations
   Terrorist tracking




                                      31
       Accident Scenario: storage &
                 retrieval
 Private Cars:
   Periodically collect images on the street (store data locally)
   Process the data and classify the event
   Create Meta-Data for event -- Summary (Type, Option,
     Location, Vehicle ID, …)
   Post it on a “distributed index”
 The police access data from distributed storage

                                          Summary
                                          Harvesting
                           - Sensing
                           - Processing




                   CRASH




                                          Crash Summary
                                             Reporting



                                                                     32
  Epidemic Posting & Harvesting
 Exploit “mobility” to create index and
  disseminate summaries
 Vehicles periodically broadcast summary of
  sensed data to their neighbors
   Data “owner” advertises only “his” own
    summaries to his neighbors
   Neighbors listen to advertisements and store
    them into their local storage
 A mobile agent (the police) harvests
  summaries from mobile nodes by actively
  querying mobile nodes
   Vehicles return all “summaries” collected so far
                                                       33
Epidemic Diffusion
- Idea: Mobility-Assist Summary Diffusion




                                        34
Epidemic Diffusion
- Idea: Mobility-Assist Summary Diffusion




                                        Keep “relaying”
                                        its summary to
                                        its neighbors


                   1) “Periodically” Relay (Broadcast)
                       its summary to Neighbors
                   2) Listen and store
                      other’s relayed summaries
                      into one’s storage                 35
Epidemic Diffusion
- Idea: Mobility-Assist Summary
Harvesting




         Sum. Rep

                      Sum. Req

          1. Agent (Police) harvests
             summaries from its neighbors
          2. Nodes return all the summaries
             they have collected so far




                                              36
Harvesting Analysis
 Metrics
   Fraction of harvested summaries F(t)
 Analysis assumption
   Discrete time analysis (time step Δt)
   N disseminating nodes
   Each node ni advertises a single summary si




                                                  37
 Harvesting Analysis-Regular Nodes

 Expected number (α) of contacts in ∆t:
   ρ : density of disseminating nodes
   v : average speed
   R: communication range

               s=vΔt
          2R



 Incremental number of summaries harvested
  by a regular node ∆Et = Et - Et-1:
   Prob. of meeting a not yet infected node is 1-Et-1/N



                                                           38
Harvesting Analysis- Agent Node
 Agent harvesting summaries from its neighbors
  (total α nodes)
 A regular node has “passively” collected so far Et
  summaries
   Probability that agent can collect a specific
    summary=Et/N
 Specific summary collected from α neighbors with
  probability 1-(1-Et/N)
 Let E*t = Expected number of summaries harvested
  by the agent




                                                       39
Harvesting Analysis - Harvesting Fraction

 Numerical analysis


                                  Area: 2400x2400m2
                                  Radio range: 250m
                                  # nodes: 200
                                  Speed: 10m/s
                                  k=1 (one hop relaying)
                                  k=2 (two hop relaying)




                                                  40
Simulation
 Simulation Setup
    Implemented using NS-2         Westwood Area

    802.11a: 11Mbps, 250m
     transmission range
    Network: 2400m*2400m
    Mobility Models
      Random waypoint
       (RWP)
      Urban map model:
       Group mobility model
       Random Merge and split at
        intersections
       Westwood map

                                                    41
Simulation
 Summary harvesting results with
  random waypoint mobility




                                    42
Simulation
 Summary harvesting results with
  urban map mobility




                                    43
Future Work
 Further investigate dependence of
  dissemination/harvesting from motion
 Enhance track models to reflect
  realistic (urban, open) scenarios
 Motion pattern characterization
   NCR (Neighborhood Change Rate)
   Fraction of “traveling buddies”, etc
 Data mining in large spatial-temporal
  databases on mobile platforms


                                           44