Redes Inalámbricas Máster Ingeniería de Computadores 2008/2009Redes Inalámbricas Máster Ingeniería de Computadores 2008/2009

Tema 5.- Redes inalámbricas Ad Hoc. Quality of Service (QoS)Tema 5.- Redes inalámbricas Ad Hoc. Quality of Service (QoS)

IntroducciónQoS in IP based NetworksQoS in MANETs Propuestas del Grupo GRC

Arquitectura DACME

QoS in MANETsQoS in MANETs

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

2

Introduction

The evolution of the Multimedia Technology and the Commercial Interest of Companies to reach civilian applications have made QoS in MANETs an unavoidable task.

QoS and Overhead are synonyms !. The idea of providing QoS in MANETs is not to extinct Overhead but to keep it as low as possible.

MANETs : 3 new problems! Dynamic Topology. Bandwidth Constrains. Limited Processing & Storing capabilities of Devices.

What happens with QoS in Wire-based Networks?. Can we port ideas / protocols to MANETs?

A Glance At QoS in Mobile Ad-Hoc Networks: http:/www.cs.ucr.edu/~csyiazti/cs260.html

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

3

The QoS Metrics

How do we measure the QoS ? Some mostly used QoS attributes

Available Bandwidth Probability of packet loss Delay variance (jitter, unpredictable delay) end-to-end delay (Accumulation of jitter along the path) Power consumption or battery charge Service coverage area

QoS Metrics can be defined in terms of one of the parameters or a set of parameters in varied proportions

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

4

QoS Definition

QoS definition “The collective effect of service performance which

determines the degree of satisfaction of a user of a service”.

The United Nations Consultative Committee for International Telephony and Telegraph (CCITT) Recommendation E.800

Video frame without QoS Support Video frame with QoS Support

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

5

Principles for QOS Guarantees (I)

Consider a phone application at 1Mbps and an FTP application sharing a 1.5 Mbps link. Bursts of FTP can congest the router and cause audio packets to be

dropped. Want to give priority to audio over FTP.

PRINCIPLE 1: Marking of packets is needed for router to distinguish between different classes; and new router policy to treat packets accordingly.

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

6

Principles for QOS Guarantees (II)

PRINCIPLE 2: provide protection (isolation) for one class from other classes. Applications misbehave (audio sends packets at a rate higher than

1Mbps assumed above). Require Policing Mechanisms to ensure sources adhere to bandwidth

requirements; Marking and Policing need to be done at the edges:

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

7

Principles for QOS Guarantees (III)

PRINCIPLE 3: While providing isolation, it is desirable to use resources as efficiently as possible.

Alternative to Marking and Policing: allocate a set portion of bandwidth to each application flow; can lead to inefficient use of bandwidth if one of the flows does not use its allocation.

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

8

Principles for QOS Guarantees (IV)

PRINCIPLE 4: Need a Call Admission Process; application flow declares its needs, network may block call if it cannot satisfy the needs . Remember: Cannot support traffic beyond link capacity

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

9

QoS in IP Based Networks

How is QoS achieved? “Over Provisioning”. Add plentiful capacity to the network.

Easy! (e.g. upgrade from 10Mb to 100Mb) Can be done gradually. But we remain at 1 service class (best effort) again.

“Network Traffic Engineering”. Make the Network more sophisticated! (e.g. Traffic Classes, Connection Admission Control, Policy Managers,…)

Reservation-based Engineering. (e.g. RSVP/IntServ, ATM) Reservation-less Engineering. (e.g. DiffServ)

– Used in today’s Differentiated Services» IPv4 TOS octect» IPv6 traffic Class octect

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

10

Integrated Services

Attempt to modify Internet service model to support diverse application requirements

Any data flow that desires better than best-effort delivery requests and reserves resources at routers along the path RSVP is the recommended reservation protocol

If insufficient resources are available, the flow is denied admission into the network

Each router Maintains reservation state for each flow Classifies every packet and decides forwarding behavior Monitors the flow to ensure that it does not consume more than the

reserved resources Advantages

Enables fine-grained QoS and resource guarantees Disadvantages

Not scalable, harder to administer

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

11

Service Interface & Call Admission

Session must first declare its QoS requirement and characterize the traffic it will send through the network

R-spec: defines the QoS being requested by receiver (e.g., rate r)

T-spec: defines the traffic characteristics of sender (e.g., leaky bucket with rate r and buffer size b).

A signaling protocol is needed to carry the R-spec and T-spec to the routers where reservation is required; RSVP is a leading candidate for such signaling protocol.

Call Admission: routers will admit calls based on their R-spec and T-spec and base on the current resource allocated at the routers to other calls.

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

12

Differentiated Services

Moves admission control and flow monitoring to the edge of the network

Edge nodes classify and mark packets to receive a particular type of service Diff Serv Code Point (DSCP)

Packet is marked in the Type of Service (TOS) in IPv4, and Traffic Class in IPv6.

Finite set of DSCPs defined

Interior nodes determine the type of service for forwarded packets based on their DSCP values

Advantages More scalable No per-flow state Easier to administer

BIG ADVANTAGE: No state info to be maintained by routers!

Disadvantages Cannot provide the same per-flow guarantees as IntServ

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

13

Edge Router/Host Functions

Classification: marks packets according to classification rules to be specified.

Metering: checks whether the traffic falls within the negotiated profile.

Marking: marks traffic that falls within profile. Conditioning: delays and then forwards, discards, or

remarks other traffic.

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

14

QoS in MANETs

A lot of work has been done in supporting QoS in the Internet, but unfortunately none of them can be directly used in MANETs because of the bandwidth constraints and dynamic network topology of MANETs.

To support QoS, the link state information such as delay, bandwidth, cost, loss rate, and error rate in the network should be available and manageable.

However, getting and managing the link state information in MANETs is very difficult because the quality of a wireless link is apt to change with the surrounding circumstances.

The resource limitations and the mobility of hosts make things more complicated.

Hard QoS guarantee is not possible in MANETs Adaptive QoS Service Differentiation

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

15

Why QoS is Hard in Mobile Ad Hoc Network?

Dynamic Network Topology Flow stop receiving QoS provisions due to path disconnections New paths must be established, causing data loss and delays

Imprecise state information Link state changes continuously Flow states change over time

No central control for coordination Error-Probe shared medium Hidden terminal problem Limited resources availability

Bandwidth, battery Life, Storage, processing capabilities Insecure medium

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

16

Effects of congestion and mobility: PSNR

Video at 10Hz 200 seconds interval

Bursty losses

Several consecutive frames lost (video freezed)

Random losses

More uniform distortion decay

degradation due to mobility

degradation due to congestion

PSNR: The phrase peak signal-to-noise ratio, often abbreviated PSNR, is an engineering term for the ratio between the maximum possible power of a signal and the power of corrupting noise that affects the fidelity of its representation.

The PSNR is most commonly used as a measure of quality of reconstruction in image compression

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

17

Effects of congestion and mobility: jitter

Congestion jitter:

relatively small

frequent variations

Mobility jitter:

very large peaks

occasional occurrences on route change

jitter is an abrupt and unwanted variation of one or more signal characteristics

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

18

Main issues

The main issues to consider to achieve good quality are: MAC level QoS: IEEE 802.11e required to differentiate

from bandwidth greedy best-effort traffic Admission control: to avoid more connections than the

MANET can handle Increase routing effectiveness: even by using layer-2

aware routing protocols such as AODV or DSR, video transmission gaps are still too large to be handled by a video codec

For video streaming Also H.264 codec tuning:

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

19

QoS in MANETs

The first QoS Model proposed in 2000 for MANETs FQMM (Flexible QoS Model for Manet QoS Signalling)

QoS Signalling INSIGNIA (in-band signalling) dRSVP(dynamic RSVP)

QoS Routing QoS enabled routing (AODV/OLSR) CEDAR(Core-Extraction Distributed Ad-hoc Routing) Ticket based Probing (distributed QoS routing)

QoS MAC IEEE 802.11e MACA/PR

(Multiple Access Collision Avoidance with Piggyback Reservation)

prioritised binary countdown (PBC) ... and

SWAN: integrated proposalMona Ghassemian, King’s College, September 2003

QoS in MANETsQoS in MANETs

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

20

FQMM

FQMM is the first QoS Model proposed in 2000 for MANETs by Xiao et al.

The model can be characterized as a “hybrid” IntServ/DiffServ Model as the highest priority is assigned per-flow provisioning. the rest is assigned per-class provisioning.

Three types of nodes: Ingress (transmit) Core (forward) Egress (receive)

The role of each nodechange according to thenode mobility

Only works with TCP traffic

1

2

5

3

4

6 7

ingress

egress

core

Mona Ghassemian, King’s College, September 2003

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

21

QoS Signalling Terminology

Signaling is used to reserve and release resources.

Prerequisites of QoS Signalling Reliable transfer of signals between routers Correct Interpretation and activation of the appropriate mechanisms

to handle the signal. It means that signaling must be understandable and implemented by the rest

of the nodes

Signaling can be divided into “In-band” and “Out-of-band” In-band: integrated in data packets Out-of-band: explicit use of control packets. Performance?

This packets should have higher priority RSVP is an example of out-of-band signaling

– Is the facto signaling protocol for IntServ

Most papers support that “In-band” Signaling is more appropriate for MANETs.

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

22

In-band VS Out-of Band Signaling

In-band Signaling, network control information is encapsulated in data packets Lightweight Not Flexible for defining new Service Classes.

Out-of-band Signaling, network control information is carried in separate packets using explicit control packets. Heavyweight signaling packets must have higher priority to achieve on time notification

=> can lead to complex systems.+ Scalability. Signal packets don’t rely on data packets+ We can have rich set of services, since we don’t need to “steal“ bits from

data packets

Source Address

Destination Address

TTL Header CheckSum

Fragment Offset

Total Length

Options Padding

Identification

Protocol

Flags

Version Hdr Len Prec TOS

32 bits

(Shaded fields are absent from IPv6 header)

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

23

INSIGNIA

INSIGNIA is the first signaling protocol designed solely for MANETs by Ahn et al. 1998. Lee, S.B., Ahn, G.S., Campbell, A.T., "Improving UDP and TCP

Performance in Mobile Ad Hoc Networks with INSIGNIA", June 2001, IEEE Communication Magazine.

Can be characterized as an “In-band RSVP” protocol. It encapsulates control info in the IP Option field (called now INSIGNIA

Option field). (IN-BAND) It keeps flow state for the real time (RT) flows. (RSVP) It is “Soft State”. The argument is that assurance that resources are

released is more important than overhead that anyway exists. (RSVP) INSIGNA tries to provide something better than best effort

service for some flows, e.g., video, voice. QoS insensitive flows can be serviced in best effort manner: e-mail QoS sensitive flows should be treated in better than best effort

manner

Mona Ghassemian, King’s College, September 2003

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

24

INSIGNIA Review

INSIGNIA is just the signaling protocol of a complete QoS Architecture. To realize a complete QoS Architecture we also need many

other components A Routing Protocol (e.g. DSR, AODV, TORA) to track changes of routes An Admission Control Module to allocate requests according to the

requested resources A Packet Scheduling Module A Medium Access Controller Module

INSIGNIA Drawbacks. Only 2 classes of services (RT) and (BE). Flow state information must be kept in mobile hosts. Georgiadis, Jacquet, and Mans proved that bandwidth

reservation on ad-hoc networks is an np-hard problem [1]

[1] “Bandwidth Reservation in Multihop Wireless Networks: Complexity and Mechanisms”. ICDCSW'04, Hachioji - Tokyo, Japan, March 2004.

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

25

QoS in MANETs, an Integrated Vision

QoS Routing QoS enabled routing (AODV/OLSR) CEDAR(Core-Extraction Distributed Ad-hoc Routing) Ticket based Probing (distributed QoS routing) Predictive Location-Based QoS Routing Protocol Bandwidth Routing Protocol Trigger-Based Distributed QoS Routing Protocol On-Demand QoS Routing Protocol QoS-Enabled Ad Hoc On-Demand Distance Vector Routing

Protocol On-Demand Link-state Multipath QoS Routing Protocol Asynchronous Slot Allocation Strategies …

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

26

QoS Routing

Routing is an essential component for QoS. It can inform a source node of the bandwidth and QoS availability of a destination node

We know that AODV is a successful an on-demand routing protocol based on the ideas of both DSDV and DSR.

We also know that when a node in AODV desires to send a message to some destination node it initiates a Route Discovery Process (RREQ).

Mona Ghassemian, King’s College, September 2003

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

27

QoS for AODV

QoS for AODV was proposed in 2000 by C. Perkins and E. Royer.

The main idea of making AODV QoS enabled is to add extensions to the route messages (RREQ, RREP).

A node that receives a RREQ + QoS Extension must be able to meet the service requirement in order to rebroadcast the RREQ (if not in cache).

In order to handle the QoS extensions some changes need to be on the routing tables

AODV current fields. Destination Sequence Number, Interface, Hop Count, Next Hop, List of Precursors

AODV new fields. (4 new fields)1. Maximum Delay, 2. Minimum Available Bandwidth, 3. List of Sources Requesting Delay Guarantees and 4. List of Sources Requesting Bandwidth Guarantees

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

28

QoS-Extensions of AODV: Basic Idea

QoS information is added to the RREQ packet

Intermediate nodes forward the RREQ only if they have sufficient resources to meet the QoS requirement

Resource information is updated in the RREQ by intermediate nodes

Destination sends resource information back to source in the RREP message

S D

RREQ

RREP

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

29

QoS for AODV - Delay

Handling Delay with the Maximum Delay extension and the List of Sources Requesting Delay Guarantees. RREQ includes delay Each node has its NODE_TRANSVERSAL_TIME

Example shows how the with the Maximum Delay extension and the List of Sources Requesting Delay Guarantees are utilized during route discovery process.

cachedelay(C->D)=50 =TraversalTime

+ delay

RREQ2delay=10ingress

Acore C

Traversal_time= 5 0

core BTraversal_time= 3 0

RREQ1delay=100

egressD

RREQ1delay=70

RREQ1delay=20

RREP1delay=0

cachedelay(B->D)=80

RREP1delay=50

RREP1delay=80

1

2

x

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

30

QoS for AODV - Bandwidth

Handling Bandwidth is similar to handling Delay requests. Actually a RREQ can include both types.

Example shows how the with the Minimum Available Bandwidth extension and the List of Sources Requesting Bandwidth Guarantees are utilized during route discovery process.

RREQ2minband=80K

cacheband(C->D)=50

ingressA

core CAvailable_Bandwidth

= 50K

core BAvailable_Bandwidth

= 100K

egressD

RREP1bandwidth=INF

cacheband(B->D)=50

RREP1bandwidth=50

2x

RREQ1min_bandwidth=10Kbps

RREQ1min_bandwidth=10Kbps

RREQ1min_bandwidth=10Kbps

RREP1bandwidth=50

min{INF,50}

1

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

31

QoS for AODV - Loosing QoS

cachedelay(C->D)=50

ingressA

core CTraversal_time= 5 0

core BTraversal_time= 3 0

egressD

cachedelay(B->D)=80

cachedelay(B->D)=80

QOS_LOSTQOS_LOST

Loosing Quality of Service Parametersif after establishment a node detects that the QoS can’t be maintained any more it originates a ICMP QOS_LOST message, to all depending nodes.== > Reason why we keep a List of Sources Requesting Delay/Bandwidth Guarantees.

Reasons for loosing QoS Parameters. Increased Load of a node. Why would a node take over more jobs that it can handle?

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

32

QoS in MANETs, an Integrated Vision

QoS MAC IEEE 802.11e Cluster TDMA MACA/PR. (Multiple Access Collision Avoidance with Piggyback

Reservation)

Prioritised binary countdown (PBC) …

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

33

... andSWAN: integrated proposal

Stateless Wireless Ad-hoc Networks intermediate nodes don’t keep per-flow or aggregate

state information differentiate real-time and best-effort traffic QoS-capable MAC not needed AIMD algorithm (+ * - like TCP window) Uses feedback information (ECN – explicit

congestion notification)

Principles: Rate control: per-hop MAC delay measurements Source-based admission control implemented in ns-2 and Linux/AODV

http://comet.ctr.columbia.edu/swan/

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

34

Propuestas del Grupo GRC: Arquitectura DACME

Previous proposals have strong requirements: All terminals must be equipped with the same software and similar

hardware All terminals must perform QoS related tasks If some of the terminals do not offer QoS support, the whole QoS

framework fails or there is severe malfunctioning

None of the previous proposals has taken into consideration that: The bandwidth reservation process is NP-hard QoS at the MAC layer is fundamental Multipath routing algorithms can offer important benefits The MANET paradigm is based on user cooperation, but in most cases

we can not force users to cooperate

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

35

Propuestas del Grupo GRC: Arquitectura DACME

Propuestas del Grupo GRC Arquitectura DACME

IP

IEEE 802.11e

IEEE 802.11g

MDSR

DAC

ME

TCP/UDP

Distributed admission control

Prioritized channel access

Multipath routing algorithm [1]

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

36

DACME - Requirements

Only two:

All stations that have IEEE 802.11e interfaces should map the IP packet's TOS to a MAC-level Access Category (basic requirement to achieve good performance)

Sources and destinations of QoS traffic should implement DACME (Distributed Admission Control for MANET Environments)

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

37

DACME - Admission control

DACME makes periodic end-to-end network measurements using probes

Intermediate stations are not aware of DACME's tasks DACME uses UDP/IP

Decisions on whether to admit, maintain or drop a QoS flow are based on DACME's periodic measurements and the QoS requirements of each specific flow

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

38

DACME architecture

1. The application registers with DACME, indicating the source and destination port, the destination's IP address and the QoS requirements

2. DACME periodically sends probes to assess available bandwidth on the path

3. The port state is set to up or down according to current network conditions

4. The packet filter module is responsible for enforcing accept/reject decisions, and also for changing the packet's TOS field if accepted

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

39

End-to-end bandwidth estimation

Is based on measurements made every 3 seconds (±0.5 s of jitter)

each probe consists of 10 back-to-back packets with the same TOS/AC as the application's packets to avoid the stolen bandwidth problem (Breslau et al., SIGCOMM 2000)

...

Δt→0

Δtrec

Source Destination

Probe reply

X

1

2

n

timeout

Probe

.....

Adjustment of these values at the source (over-estimation)

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

40

Performance evaluation

General simulation setup: ns-2 discrete event simulator Radio interfaces are IEEE 802.11g/e enabled Scenarios are sized 1900x400 m2 and composed by 50

nodes Radio range is of 250 meters (4 hops between nodes on

average) Nodes move according to the random way-point model at

a constant speed of 5 m/s Comparison between DSR & MDSR routing protocols Simulation time is of 300 seconds for each experiment DACME source/destination pairs have a DACME agent

attached

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

41

Traffic

4 Video sources and 3 Voice sources regulated by DACME Video sources generate CBR traffic at 1 Mbit/s in the Video AC Voice sources:

VoIP streams simulated using a Pareto On/Off distribution both burst and idle time set to 500 ms shaping factor used is 1.5, average data rate is of 100 kbit/s

Sources are turned off in the same order they were turned on 4 background traffic sources

Traffic is negative-exponentially distributed Variable traffic loads; load share per AC is: 50% to the Video AC, 25%

to Best-effort AC, 25% to Background AC These sources are active all the time

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

42

Performance in terms of throughput/losses

On average the throughput of DACME-regulated sources is much more stable (always close to the source data rate of 1 Mbit/s)

Voice sources do not generate constant data-rate traffic In terms of packet losses we achieve very significant

improvements

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

43

Performance in terms of end-to-end delay

In terms of average end-to-end delay, DACME allows achieving much lower values than its non-DACME counterpart

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

44

Routing overhead and traffic acceptance rate

In terms of routing overhead, DACME reduces it by avoiding routing collapse situations

In terms of traffic acceptance rate: high data-rate sources (video) are more penalized

Red

es I

nalá

mb

ricas

M

IC

20

08/2

009

45

Conclusions and future work

We introduced a new paradigm of QoS architecture for MANETs based on distributed admission control that is able to adapt to the different constrains of MANET environments

Simulation results show that DACME: Improves the support of multimedia applications by achieving more

stable throughput, fewer packet losses and reduced end-to-end delay Does not misbehave when combined with a multipath routing

protocol (MDSR) Promotes routing stability and efficient usage of the radio channel

In the future we plan to develop a version of DACME for the Linux operating system to deploy an IEEE 802.11e-based real-life testbed In linux system DACME can be implemented using Iptables