Identifying High Throughput Paths in 802.11 Mesh Networks : A Model-based Approach

1
Identifying High Throughput Paths in 802.11 Mesh
Networks A Model-based Approach
Theodoros Salonidis (Thomson) Michele Garetto
(University of Torino) Amit Saha (Tropos)
Edward Knightly (Rice University)
2
Hot-spot wireless networks
Internet
Internet
802.11
802.11
Internet
Internet
Internet
802.11
802.11
802.11

• Cellular-like high-speed wireless data networks
• Use 802.11 for user access and wired Internet for
  backbone

3
Multi-hop wireless mesh networks
Internet
802.11
802.11
802.11 wireless links
802.11
802.11
802.11

• Aim Low-cost / high-speed wireless access
• Use 802.11 for both user access and backbone
• Scale Neighborhood to city-wide, US/Europe/Asia

4
Multi-hop wireless mesh networks
Internet
802.11
802.11
802.11 wireless links
802.11
802.11
802.11

• Fact 802.11 CSMA MAC protocol is used for both
  user access and backbone
• Problem Severe throughput imbalances and
  starvation

5
Our contributions

• Analytical model
• Predict per-flow throughput in arbitrary
  topologies employing 802.11 MAC protocol.
• Explain the origin of starvation in CSMA-based
  multi-hop wireless networks
• Solution
• High-throughput mesh routing

6
Roadmap

• Overview of multi-hop 802.11 model
• Technique for available bandwidth computation
• Comparison of existing loss-based routing metrics
  with new routing metric that directly computes
  high-throughput paths

7
Analytical model

• The channel view of a node

Nodes transmission collides
channel busy due to activity of other nodes
Nodes transmission is successful
idle slot


t

• Modeled as a renewal-reward process

P event Ts occurs
Throughput (pkt/s)
Average duration of an event (s)
8
Analytical model

• Define

probability that the node
sends a packet
conditional collision probability
conditional busy channel probability

• Event probabilities

Success
Busy channel
Idle
Collision


t
9
Analytical model

• Throughput formula (saturated link)

10
Available bandwidth estimation

• Inter-flow step at each node
• Use measured values of fB and p on adjacent links
• Compute additional input rate needed to saturate
  each link
• Intra-flow step
• Clique-based formulation to capture bandwidth
  sharing among links within the path

11
Model validation

• Topology
• Chaska.net
• 196 APs / 14 GWs
• Simulation setup
• 802.11b, single channel
• Download/Upload traffic
• Load gateways 2Mbps

12
Model validation
Chaska download scenario
Chaska upload scenario

• Good match between model available BW and
  achieved throughput

13
Loss-based (LB) routing metrics

• ETX (MIT)

• ETT (Microsoft)

• IRU (UIUC)

• LB metrics are load-sensitive and depend only on
  packet loss probability p

14
Single link performance
15
LB metrics can pick suboptimal paths
G1
?
A
G2
B
C
16
AVAIL vs. LB metrics

• AVAIL model-based routing metric
• Aim
• Compare AVAIL with LB metrics (ETX, ETT and IRU)
• Routing protocol
• LQSR link state, source routing
• Each node periodically broadcasts measured fB, p
• Each node uses modified Dijkstra to compute AVAIL
• Simulation setup
• 100 initial UDP upload flows (pick min-hop
  gateways)
• One incoming UDP flow (50 random samples)
• Rate limiting
• For all metrics, incoming flow rate-limited based
  on model

17
Chaska comparison

• Max gateway load 2Mbps
• LB metrics AVAIL Tput on average

18
Manhattan topology

• Topology
• 14x14 / 4-neighbor
• 196 APs / 10 GWs
• Simulation setup
• 802.11b, single channel
• Upload traffic
• Load gateways (30-100) x maxload

19
Manhattan comparison

• Max gateway load 3Mbps

20
Manhattan comparison

• Max gateway load 4Mbps

21
Conclusions

• Analytical model accurately predicts available
  bandwidth
• Busy time crucial for high throughput routing
• LB metrics can pick suboptimal/starving paths
• Topologies that allow spatial reuse and longer
  paths yield highest gains