Identifying High Throughput Paths in 802.11 Mesh Networks : A Model-based Approach Theodoros Salonidis (Thomson) Michele Garetto (University of Torino)

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

InternetInternet Internet

– Cellular-like high-speed wireless data networks

– Use 802.11 for user access and wired Internet for backbone

802.11

802.11

802.11

802.11 802.11

3

Internet

– Aim: Low-cost / high-speed wireless access

– Use 802.11 for both user access and backbone

– Scale: Neighborhood to city-wide, US/Europe/Asia

802.11 wireless links802.11

802.11

802.11

802.11 802.11

Multi-hop wireless “mesh” networks

4

Multi-hop wireless “mesh” networks

Internet

– Fact: 802.11 CSMA MAC protocol is used for both user access and backbone

– Problem: Severe throughput imbalances and starvation

802.11 wireless links802.11

802.11

802.11

802.11 802.11

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

The “channel view” of a node:

… …

Node’s transmission is successful idle slot

Node’s transmission collides

t

channel busy due to activity of other nodes

Modeled as a renewal-reward process

Throughput (pkt/s) =P [event Ts occurs]

Average duration of an event (s)

Analytical model

8

… …t

Define: = probability that the node sends a packet

= conditional collision probability= conditional busy channel probability

Success Idle Collision Busy channel

Event probabilities

Analytical model

9

Throughput formula (saturated link)

General throughput formula

)(,, ltxBlll fpfT

Input ratePacket loss

probability

Fraction of busy time

Analytical model

10

Available bandwidth estimation

1

25

1

100

1

50

1

Inter-flow step at each node

– Use measured values of fB and p on adjacent links

– Compute additional input rate needed to saturate each linkIntra-flow step

– Clique-based formulation to capture bandwidth sharing among links within the path

1 2 43

50 pkt/sec 100 pkt/sec 25 pkt/sec 20 pkt/sec

1

20

1

25

1

100

1

Path BW = min ( ) = 10 pkts/sec ,

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

ll p

ETX

1

1

l

lll B

SETXETT

lll NETTIRU

14

Single link performance

Large deviation for high busy time!

LB metrics Tput– Linear on p

ll pTT 1max

Model Tput– Non-linear on p

– Linear on fB

)(,, ltxBlll fpfT

15

LB metrics can pick suboptimal pathsAG1 B

C

G2?

Load C->G1

Achievable G1

Achievable unused G2

LB metrics Tput loss

16

AVAIL vs. LB metricsAVAIL: model-based routing metricAim

– 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 AVAILSimulation 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

AVAIL metric achieves

x1.5 gain on average

20

Manhattan comparison

Max gateway load: 4Mbps

AVAIL metric achieves

x2.4 gain on average

LB metrics

starve!

21

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

Conclusions