WPMC 2003 Yokosuka, Kanagawa (Japan) 21-22 October 2003

Department of Information EngineeringUniversity of Padova, ITALY

On Providing Soft-QoS in On Providing Soft-QoS in

Wireless Ad-Hoc NetworksWireless Ad-Hoc Networks

{andrea.zanella, daniele.miorandi, silvano.pupolin}@dei.unipd.it

Andrea Zanella, Daniele Miorandi, Silvano Pupolin, Paolo Raimondi

WPMC 2003, 21-22 October 2003

Special Interest Group on NEtworking & Telecommunications

WPMC 2003 Yokosuka, Kanagawa (Japan) 21-22 October 2003

Motivations

Ad-hoc networks are a valuable solution to

Extend in a multi-hop fashion the radio access to wired networks

Interconnect wireless nodes without any fixed network structure

In these contexts, providing QoS is a key issue

audio/video streaming

interactive games

multimedia

A possible QoS support method

QoS-routing & Call-Admission-Control (CAC) mechanisms

Constrained Shortest Path Routing Problem (NP-complete)

MAC-layer Resource Reservation (MRR) and scheduling strategies

WPMC 2003 Yokosuka, Kanagawa (Japan) 21-22 October 2003

Hard & Soft QoS

Widely used in wired

networks

Integrated Services: flow

based (RSVP)

Differentiated Services:

class based

Suitable for wireless

networks

Applications may work even

if, for short periods of time,

QoS requirements are not

satisfied

Deals with limited bandwidth

and radio channel

Hard-QoS Soft-QoS

WPMC 2003 Yokosuka, Kanagawa (Japan) 21-22 October 2003

Aim of the study Reference network scenario

Low-profileLow-profile multi-hop wireless networks Intermediate nodes capable of basic functionalities

• Routing – Link monitoring – Basic computation

Border nodes capable of rather complex functionalities

• Call Admission Control (CAC) – MAC layer Resource Reservation (MRR)

Goal Providing Soft-QoSSoft-QoS support over low-profile multi-hop networks

Define Soft QoSSoft QoS parameters

Define distributed statistical CACdistributed statistical CAC

Define statistical MAC-layer Resource Reservation (MRR) mechanismstatistical MAC-layer Resource Reservation (MRR) mechanism

Modify AODVAODV in order to support Soft-QoS routing

WPMC 2003 Yokosuka, Kanagawa (Japan) 21-22 October 2003

What’s Soft-QoS?

Soft-QoS definitionSoft-QoS definition

WPMC 2003 Yokosuka, Kanagawa (Japan) 21-22 October 2003

QoS parameters required per link

Minimum peak band: Br

End-to-End Delay: Dr

Soft QoS parameter: Target Satisfaction indexTarget Satisfaction index

r = percentage of pcks expected to satisfy QoS

constrains

r = 1 hard QoS (or “wealthywealthy” clients)

r = 0 pure best-effort (or “poorpoor” clients)

Soft-QoS parameters

WPMC 2003 Yokosuka, Kanagawa (Japan) 21-22 October 2003

Call Admission Control

Distributed CAC Distributed CAC mechanism mechanism

WPMC 2003 Yokosuka, Kanagawa (Japan) 21-22 October 2003

Path Service Levels

Path: P = (p1,…, pN)

Service levels:

Path Peak Bandwidth

Path Delay

Pp

rprP

j

jBbBB PrPr}{min

jj

pPp

P bB

Pp

pP

j

jdD

Pp

dD

j

jpPsHsH )()(

WPMC 2003 Yokosuka, Kanagawa (Japan) 21-22 October 2003

CAC & Gaussian approx Path is feasible if

Bandwidth constrained requests

Delay constrained requests

This would require the collection of the complete statistics of link bandwidth and delay… but we may (always?) resort to the Gaussian approximationGaussian approximation

Statistics are univocally determined by mean and standard deviation values

Bandwidth constrained requests

Delay constrained requests

rrP BB Pr

rrP DD Pr

rPp b

br

j jp

jpmB

Q

rD

Dr

P

PmD

Q

1

WPMC 2003 Yokosuka, Kanagawa (Japan) 21-22 October 2003

MRR

Statistical MAC-layer Statistical MAC-layer Resource ReservationResource Reservation

WPMC 2003 Yokosuka, Kanagawa (Japan) 21-22 October 2003

Bandwidth Margins

Bandwidth-constrained requests

Resource bounds Minimal residual resources that should be guaranteed to

preserve QoS levels of accepted connections

r

jp

Pp b

br

b

brb

k

k kp

kp

j

j

j

mBQ

mBQm

ˆ such that ˆ

j

j

jjb

brrbrb

mBQQBm

ˆ

ˆ 1

WPMC 2003 Yokosuka, Kanagawa (Japan) 21-22 October 2003

Delay Margins

Delay-constrained requests

Extra-delay margin is computed for the entire path

Each link along the path is assigned a fraction of the extra delay time inversely proportional to the average link delay

ˆ1ˆ 1 QDmPP DrD

jj

k

j

jj DD

Pkd

ddd mm

m

mmm

ˆ1

1ˆ

rD

DrD

P

P

P

mDQm

ˆ1 such that ˆ

WPMC 2003 Yokosuka, Kanagawa (Japan) 21-22 October 2003

Maximum Sustainable traffic

Before a connection request is accepted the tightest

resource margins among the nodes along the path are

made available at the source

From these bounds, the source derives the maximum

sustainable traffic rate, i.e., the maximum traffic that can

be injected into the network without violating the QoS

agreements of the connections already established

WPMC 2003 Yokosuka, Kanagawa (Japan) 21-22 October 2003

How to create a path

Soft-QoS routing Soft-QoS routing algorithmalgorithm

WPMC 2003 Yokosuka, Kanagawa (Japan) 21-22 October 2003

Path creation & maintenance Soft-QoS routing is largely inspired to AODV

Each Route Request (RREQ) packet gathers statistical information on the minimum bandwidth and maximum delay along that portion of the path

RREQ is propagated only whether bandwidth request is satisfied

The destination node back propagates a Route Reply (RREP) packet along the selected path

RREP acquaints intermediate nodes with new resource bounds and updates maximum sustainable traffic rate limit

Source node is required to respect the maximum sustainable traffic rate limit or to refuse the connection

WPMC 2003 Yokosuka, Kanagawa (Japan) 21-22 October 2003

Simulation Results

Simulation of Soft-QoS Simulation of Soft-QoS routing algorithmrouting algorithm

WPMC 2003 Yokosuka, Kanagawa (Japan) 21-22 October 2003

Simulation Scenario

Bluetooth Scatternet Round Robin Polling

Gateways spend 50 slots in each piconet

Poisson packets arrival process Mixed packet formats with average length of 1500 bits

Delay-constrained requests

WPMC 2003 Yokosuka, Kanagawa (Japan) 21-22 October 2003

Gaussian Approximation Local slave-to-slave connections

in each piconet Data rate=9.6 Kbit/s 1 hop

6 hops

Gaussian approx is fairly closefairly close to empirical delay CDF

Gap increases for long-distance and high traffic connection

WPMC 2003 Yokosuka, Kanagawa (Japan) 21-22 October 2003

Simulation setup Target connection c1

Dr = 50 ms

r = 0.2 r = 20 kbit/s

Target connection c2

Dr = 200 ms

r = 0.9 r = 30 kbit/s

Target connection c3

Dr = 200 ms

r = 0.9 r = 20 kbit/s

Target connection c4 Dr = 50 ms r = 0.2 r = 60 kbit/s

Transversal connections Starting after 20 s, last for 10 s On average 1 request/s Random source, destination & QoS

requests Rate: 520 kbit/s

WPMC 2003 Yokosuka, Kanagawa (Japan) 21-22 October 2003

Satisfaction & Delay dynamics Satisfaction Delay

WPMC 2003 Yokosuka, Kanagawa (Japan) 21-22 October 2003

Conclusions We have proposed a basic Soft QoS routing algorithm for low-profile

ad hoc networks

Provides Soft-QoS guarantees

Requires

basic nodes’ functionalities

statistical link state monitoring (mean and standard deviation)

Does not require

service differentiation

static resource reservation

Drawbacks

Lower resource utilization

Higher rate of connection request rejection

WPMC 2003 Yokosuka, Kanagawa (Japan) 21-22 October 2003

Department of Information EngineeringUniversity of Padova, ITALY

On Providing Soft-QoS in Wireless Ad-Hoc NetworksOn Providing Soft-QoS in Wireless Ad-Hoc NetworksAndrea Zanella, Daniele Miorandi, Silvano Pupolin, Paolo Raimondi

Questions?

WPMC 2003 Yokosuka, Kanagawa (Japan) 21-22 October 2003

Extra Slides…

Spare SlidesSpare Slides

WPMC 2003 Yokosuka, Kanagawa (Japan) 21-22 October 2003

Statistical Resource Reservation

Bandwidth-constrained

Delay-constrained Extra-delay margin given to

each link along the path is inversely proportional to the mean link delay

Resource bounds Minimal residual resources that should be guaranteed

to preserve QoS levels of accepted connections

j

j

jjb

brrbrb

mBQQBm

ˆ

ˆ 1

ˆ1ˆ 1 QDmPP DrD

jj

k

j

jj DD

Pkd

ddd mm

m

mmm

ˆ1

1ˆ

Actual SatisfactionResource bounds