Modeling Per-flow Throughput and Capturing Starvation in CSMA - - PowerPoint PPT Presentation

▶
modeling per flow throughput and capturing starvation in
SMART_READER_LITE
LIVE PREVIEW

Modeling Per-flow Throughput and Capturing Starvation in CSMA - - PowerPoint PPT Presentation

Mar 01, 2024 396 likes •699 views

Modeling Per-flow Throughput and Capturing Starvation in CSMA Multi-hop Wireless Networks Michele Garetto Theodoros Salonidis Edward W. Knightly Rice Networks Group http://www.ece.rice.edu/networks Modeling Per-flow Throughput and Capturing

slide-1
SLIDE 1

Rice Networks Group http://www.ece.rice.edu/networks

Michele Garetto Theodoros Salonidis Edward W. Knightly

Modeling Per-flow Throughput and Capturing Starvation in CSMA Multi-hop Wireless Networks

slide-2
SLIDE 2

Rice Networks Group http://www.ece.rice.edu/networks

Michele Garetto Theodoros Salonidis Edward W. Knightly

Modeling Per-flow Throughput and Capturing Starvation in CSMA Multi-hop Wireless Networks

slide-3
SLIDE 3

100 200 300 400 500 600 700 800 900 1000 100 200 300 400 500 600 700 800 900 1000 Y (meters) X (meters)

Example : 50 nodes

slide-4
SLIDE 4

100 200 300 400 500 600 700 800 900 1000 100 200 300 400 500 600 700 800 900 1000 Y (meters) X (meters)

50 tx-rx pairs (link flows)

Example : 50 nodes

Saturated traffic 802.11 DCF (CSMA/CA) Perfect channel

slide-5
SLIDE 5

100 200 300 400 500 600 700 800 900 1000 100 200 300 400 500 600 700 800 900 1000 Y (meters) X (meters)

Example : 50 nodes

Single cell

Sensing range

slide-6
SLIDE 6

Example : 50 nodes

10 20 30 40 50 5 10 15 20 25 30 35 40 45 50

Throughput (pkt/s) Node ID

Single cell

All flows receive equal throughput

slide-7
SLIDE 7

100 200 300 400 500 600 700 800 900 1000 100 200 300 400 500 600 700 800 900 1000 Y (meters) X (meters)

Sensing Range = 400m

100 200 300 400 500 600 700 800 900 1000 100 200 300 400 500 600 700 800 900 1000

Example : 50 nodes

slide-8
SLIDE 8

Example : 50 nodes

Throughput (pkt/s) Rank

50 100 150 200 250 5 10 15 20 25 30 35 40 45 50

A few rich flows Many starving flows !

slide-9
SLIDE 9

50 100 150 200 250 5 10 15 20 25 30 35 40 45 50 55

ideal channel fading + capture Throughput (pkt/s) Rank

Example : 50 nodes

slide-10
SLIDE 10

Our contributions

Develop an analytical model to compute

per-flow throughput in arbitrary network topologies employing 802.11 DCF

Explain the origin of starvation in CSMA-

based multi-hop wireless networks

Propose metrics to quantify starvation due

to the MAC protocol operation

slide-11
SLIDE 11

Related work on CSMA models

Models for single-cell networks (WLANs)

Leverage symmetric channel state Accurately capture carrier sense, Binary Exponential

Backoff, RTS/CTS (e.g. [Bianchi00])

Models for multi-hop networks

Assumption

Throughput proportional to number of interferers

[Boorstyn87, Carvalho04, Kar05]

Cannot capture the CSMA “disproportionalities”

  • r predict zero throughput
slide-12
SLIDE 12

Our approach

Decoupling technique

Describe behavior of each node based on its private view of

the channel state.

Throughput expression

Express throughput of each node as a function of its

Sensed fraction of busy time (b, Tb) Collision probability p

Basic iteration

Compute p, (b,Tb) variables of each node subject to the

current variables of other nodes.

Iterative solution

Perform basic iteration until convergence

slide-13
SLIDE 13

Analytical model

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)

slide-14
SLIDE 14

Analytical model

Event probabilities:

… … t

Define:

= probability that the node sends a packet

= conditional collision probability = conditional busy channel probability

Success Idle Collision Busy channel

slide-15
SLIDE 15

Unknown variables (different for each node)

Collision probability Busy channel probability Average busy time

Analytical model

  • Throughput formula:

fbianchi(.)

Deterministic decreasing function of p Captures Binary Exponential Backoff

slide-16
SLIDE 16

How can collision probability p be disproportionately large ?

b B a A The “information asymmetry” scenario

37 pkts/sec 446 pkts/sec ( = 0) ( = 0.85)

Flow A->a starves

slide-17
SLIDE 17

…

t

Why is collision probability p disproportionately large ?

…

RTS

?

View of A View of B

b B a A The “information asymmetry” scenario

37 (pkts/sec) 446 (pkts/sec) ( = 0) ( = 0.85)

Flow A->a starves due to high packet loss Starvation cause: A contends randomly B knows when to contend

slide-18
SLIDE 18

How can busy channel product bTb be disproportionally large ?

a A b B C c

The “flow-in-the-middle” scenario

30 449 pkts/sec 449

Flow A->a starves

( = 0) ( = 0) ( = 0)

No packet losses

slide-19
SLIDE 19

Why is busy channel product bTb disproportionally large ?

Channel view of A:

TxOp for A

30

a A b B C c

( = 0) ( = 0) ( = 0) 449 449

The “flow-in-the-middle” scenario

Starvation cause: A senses busy medium for a very long time Flow A->a starves No packet losses

B B B B C C C C

slide-20
SLIDE 20
  • Challenge
  • Not all neighbors of A are mutually

within range and their activities are inter- dependent.

Computation of busy channel parameters (b,Tb) for flow Aa

A

  • Clique computation
  • Find minimum number of maximal

cliques M covering all neighbors

slide-21
SLIDE 21

Computation of busy channel parameters (b,Tb) for flow Aa

A

1 4 2 7 6 5 3 M=3

  • Challenge
  • Not all neighbors of A are mutually

within range and their activities are inter- dependent.

  • Clique computation
  • Find minimum number of maximal

cliques M covering all neighbors

  • Virtual nodes (VN) graph
  • VN = set of non-empty clique

intersections

  • VN Graph: Connect two VNs if they

share at least one clique

slide-22
SLIDE 22

Computation of busy channel parameters (b,Tb) for link Aa

  • Challenge
  • Not all neighbors of A are mutually

within range and their activities are inter- dependent.

  • Clique computation
  • Find minimum number of maximal

cliques M covering all neighbors

  • Virtual nodes (VN) graph
  • VN = set of non-empty clique

intersections

  • VN Graph: Connect two VNs if they

share at least one clique

  • Computation of busy period
  • Find the aggregate busy time around

node i based on VN activities

1 4 2 7 6 5 3

slide-23
SLIDE 23

Computation of collision probability p for link Aa

1) Coordinated losses: Pco

p = 1 – (1-Pco)(1-Pia)(1-Pnh)(1-Pfh)

4 classes of packet loss due to link Bb

b B a A b B a A b B a A b B a A

2) Information Asymmetry: Pia 3) Near Hidden terminals: Pnh 4) Far Hidden terminals: Pfh

slide-24
SLIDE 24

Network solution

Basic iteration

Compute p, (b,Tb) of each node subject to the variables of

  • ther nodes.

Network solution

Multivariate system of coupled non-linear equations Perform basic iteration until convergence

Model features

Incorporates all starvation effects due to CSMA MAC Can analyze arbitrary topology Predicts individual flow throughput Supports non-saturated flows

slide-25
SLIDE 25

Model vs Sim – 50-nodes example

50 100 150 200 250 300 5 10 15 20 25 30 35 40 45 50

sim model Throughput (pkt/s) Rank

slide-26
SLIDE 26

Measuring Starvation

Objectives

Capture how individual flows are treated by different

solutions

Distinguish between imbalance due to topology

(number of contenders) and starvation due to the MAC protocol

Reference system: Slotted Aloha

Starvation structurally eliminated

slide-27
SLIDE 27

Conclusions

Multi-hop wireless networks employing 802.11 (or

  • ther variants of CSMA) are subject to severe

starvation (under high load)

This is a fundamental problem CSMA due to lack of

coordination between out-of-range transmitters

We developed an analytical model to predict per-

flow throughput in arbitrary topologies and characterize starvation

slide-28
SLIDE 28

Thank you

For more information: www.ece.rice.edu/~thsalon