Interference Considerations for QoS in MANETs

1
Interference Considerations for QoS in MANETs

• Rajarshi Gupta, John Musacchio, Jean Walrand
• guptar, musacchj, wlr_at_eecs.berkeley.edu
• University of California, Berkeley

2
Why Interference is critical

• In wired networks, all links may be used
  simultaneously
• In MANET, neighboring links interfere
• Interference Range (Ix) gt Transmission Range (Tx)
  
• For simulations
• Transmission range 500m
• Interference range 1 km

3
Overview

• Previously assumed approximate models
• No interference across clusters
• Only one hop interference
• New Contribution
• Model MANET with accurate interference
  considerations
• 802.11b MAC protocol
• Interference based on distance
• Heuristic QoS algorithms incorporating
  interference effects
• Simulation study to validate theoretical models

"Adaptive Quality of Service for a Mobile Ad Hoc
Network, A. Dimakis, L. He, J. Musacchio, H-S W.
So, T. Tung, and J. Walrand, MWCN 2003. "A
Wireless Overlay Network with QoS Capabilities,
E. Magana, D. Morato, H.W. So, B. Hodge, J.
Walrand, and P. Varaiya, Technical Report.
4
Conflict Graph

• Interference between links in graph G may be
  modeled as Conflict Graph CG
• Link from node i to node j in G gt vertex Lij in
  CG
• Edge in CG between Lij and Lpq iff Lij and Lpq
  interfere with each other
• Incorporates protocol versions
• With/out RTS-CTS (simulations only without)
• Consideration for MAC-layer acknowledgements
• Two links (i.e. vertices in CG) can not be active
  simultaneously if there is a edge connecting them

"Impact of Interference on Multi-hop Wireless
Network Performance, K. Jain, J. Padhye, V. N.
Padmanabhan, and L. Qiu, ACM Mobicom 2003.
5
Ideal Solution

• Goal
• Maximize concurrent transmissions
• Schedule many non-interfering links
• Solution
• Identify maximal sets of non-neighboring links,
  i.e Independent Sets in the C.G
• Schedule the Independent Sets s.t. the QoS
  requirements are met for flows
• Very hard problem (even if centralized)
• Finding all independent sets itself is NP-hard
• Then need to appropriately schedule

A New Model for Packet Scheduling in Multihop
Wireless Networks, H. Luo, S. Lu, and V.
Bhargavan, ACM Mobicom 2000.
6
Alternative Solution Cliques

• Clique Complete Subgraph
• Maximal Clique Clique not a subset of any other
• Only one vertex in a clique may be active at once
• Capacity in ad-hoc networks closely related to
  cliques in CG

Maximal Cliques ABC, BCEF, CDF
7
Proposed Clique-based Mechanism

• Objectives
• Fully distributed processing
• Functions only with localized information
• Dynamic
• Computationally efficient i.e. quick
• Can work (less accurately) even with incomplete
  information
• Heuristic mechanism

8
State Information Exchange

• All nodes have GPS to know their position
• Nodes need to know about all their interference
  neighbors
• Their locations
• Allocated flows at each neighbor
• Need message exchanges between interference
  neighbors
• Usually available in local neighborhood
• Works with incomplete information, but may yield
  sub-optimal decisions
• Each node has the logical information to compute
  its CG subgraph, but explicit computation not
  required

9
Computing Cliques

• General algorithms take exponential time
• Propose faster heuristic algorithm
• Key observations for an interference CG
• All links sharing cliques with this link must lie
  within a radius of Ix (interference range)
• Links that together form a clique must all lie
  within a diameter Ix

10
Heuristic Clique Algorithm

• Use a disk of radius Ix/2 to scan a disk of
  radius Ix around link
• Each position of scanning disk generates a clique
  
• Shrink set of cliques by remembering previous
  clique and checking containment
• Can further shrink to set of maximal cliques

• Time taken to generate cliques that the link
  belongs to
• 1 sec to get heuristically shrunk set of cliques
• lt15 sec to shrink to set of maximal cliques

11
Theoretical Result

• Unfortunately, capacity constraints based on
  cliques are not sufficient
• Only work for Perfect Graphs
• Need a scaling factor of
  
• for sufficiency
• Flows that satisfy scaled clique constraints have
  a realizable schedule

Clique constraints suggest a rate of 0.5 per
link But only 0.4 per link is achievable
Graph Imperfection I, S. Gerke and C.
McDiarmid, Journal of Combinatorial Theory,
Series B, vol. 83 (2001), pp. 58-78.
12
Complete Distributed Mechanism

• Local link state exchange position, flow
• Distributedly compute all maximal cliques
• Recompute upon topology change
• Requested flow (rate path) checked by all nodes
  in neighborhood of path
• Check allocated and requested flows against
  clique constraints scaled by 0.46
• Admit flows if satisfied

13
Visualization of Algorithm

• Plot ad-hoc nodes and links
• Color of a link
• denotes allocated resources on link,
• considering interference over cliques,
• expressed as of theoretical capacity
• Allocated flows
• paths shown in gray
• bandwidths shown in list

14
OPNET Simulation Model
15
Comparing Model with Simulation

• X-axis minimum spare capacity amongst all
  cliques
• Y-axis percentage of traffic received
• Blue Average over all flows
• Red Worst amongst all flows
• Each point indicates a simulation run (some runs
  are non-uniform)
• Vertical bars indicate spare capacities of
  -2, 5 and 10

16
Received vs Sent Rates

• All flows have the same sending rate
• X-axis average rate of sent traffic
• Y-axis average rate of received traffic
• Vertical lines show theoretical capacity limits
  predicted by clique constraints

-- 3 Flows -- 4 Flows -- 5 Flows
Clique Predicted Limit 3 Flows
Clique Predicted Limit 4 Flows
Clique Predicted Limit 5 Flows
17
Next Phase of Work

• Make further use of interference knowledge
• Distributed QoS routing algorithm for a general
  MANET
• To be used also for distributed intra-cluster
  routing in a clustered MANET
• Incorporate mobility in simulations
• Handle multiple classes of service