“women” project final meeting trento, feb. 12th, 2008 1 unitn: description of work fabrizio...

“WOMEN” Project Final Meeting

Trento, Feb. 12th, 2008 1

UNITN:Description of Work

Fabrizio GranelliDISI – University of [email protected]

“WOMEN” Project Final MeetingTrento, Feb. 12th, 2008 2

UNITN role in a slide Goal: “Performance Enhancement of Wireless

Mesh Networks and Testbed Implementation”

WP1 WP3

T3.4 Stream Control WP4

T4.5 Network Capacity Analysis WP5 (Resp.)

T5.6 (Testbed) (Resp.) T5.7 (Trials) (Resp.)



UNITN:Research Activities


P ay lo adI P

Head er

T3.4 “Stream Control” PAC-IP

Concatenate IP packets into a single link layer payload Share wireless overhead by entire group of packets

P ay lo adI P

Head er

Sender Receiver

P HYHead er

LLHead er

P ay lo adI P

Head er

P ay lo adI P

Head er

Link

Physical

Network


PAC-IP: Possible Implementations

I P la y e r

PA C -I PC on c ate n ator

L in k L a y e r

D e st 1

D e st 2

D e stN

+

I n t e r f a c eQ ue ue

T im e r s

M e diu m -bu s yIn dic ator

I P la y e r

PA C -I PC on c ate n ator

L in k L a y e r

D e st 1

D e st 2

D e stN

+

I n t e r f a c eQ ue ue

T im e r s


Evaluation Results Simulation

Ns-2 network simulator

0

1

2

3

4

5

6

7

0 300 600 900 1200 1500Packet size (Bytes)

Thr

ough

put

(Mbp

s)

MAC 802.11bSimulationsSingle-flow testbedMulti-flow testbed

No need to change Link or Transport layers

Good performance for small (< 750 bytes) packets

Testbed Modified Orinoco Silver

card’s driver Iperf traffic generator

ConcatenationON

ConcatenationOFF


TCP Performance Enhancement:TCP LogWestwood+ Design Objectives

To be more aggressive than Standard TCP to guarantee equal or better throughput and utilization

Fast increase for low window values while accurately approach the available pipe capacity

Small sensitivity with respect to RTT

TCP LogWestwood+ combines smart TCP Westwood window decrease with Logarithmic increase function

Ref.: D. Kliazovich, F. Granelli, and D. Miorandi, "TCP Westwood+ Enhancements for High-Speed Long-Distance Networks,“ ICC'06.


TCP LogWestwood+

Logarithmic

Linear AIMD(Standard TCP)

On packet loss (3 dup ACKs)

= * 2, if W < Wmax

= / 2, if W ≥ Wmax

≥ 2 (initial value)0 5 10 15 20Acknowledgements

Co

ng

est

ion

Win

do

w

TCP LogWestwood+TCP NewReno

Slow Start Congestion Avoidance

Logarithmic Linear

Window Control during congestion avoidance Wmax – window size before the last packet loss

occurred


TCP LogWestwood+:Performance Window Evolution

Physical Capacity

Total Capacity (+Buffers)

2Total Capacity

1000 1005 1010 1015 1020 1025 1030 1035 1040 1045 1050300

350

400

450

500

550

600

650

700

750

time (s)

Co

ng

est

ion

Win

do

w (

pa

cke

ts)

LogWestwood+Westwood+NewReno


Quantifying Cross-layer Design

Quantifying the effect of potential cross-layer interactions is very important to systematically relate such interactions to system outcomes to quantify the decision to take such interactions into account

We propose to quantify cross layer interactions by defining factors (parameters) and effects (measurements) across layers in a way that is common in system science and operations

research

Ref.: F. Granelli, M. Devetsikiotis, “Designing Cross-Layering Solutions for Wireless Networks: a General Framework and Its Application to a Voice-over-WiFi Scenario,” CAMAD 2006.


Quantifying Cross-layer Design Cross-Layer Sensitivity and Performance

Optimization

020

4060

80100

0

5

10

150

0.02

0.04

0.06

0.08

pkt ratedatarate

d2N

*/dD

dT

0.02

0.03

0.04

0.05

0.06

0.07

DTT

eDTTD

N DT

lnln00852.0ln0372.0278.0

7774.0 lnln0923.08143.05966.0*2


To define and validate an analytical model for connectivity and capacity of Wireless Mesh Networks, given design parameters n, r, l.

Traffic from/to mesh base station ≠ ad-hoc scenario

Network Capacity Analysis: goal


Problem Statement• Single-cell analysis

→ squared area

→ m = 1

→ interfering nodes

• Traffic to/from BS

• Hyp.: → IEEE 802.11 CSMA/CA – RTS/CTS*

→ MFR routing

→ r = 250 [m]

* → “Performance Analysis of the IEEE 802.11 Distributed Coordination Function”, by G. Bianchi (JSAC, March 2000)


Connectivity analysis

No. offorwarding nodes*

Poisson distribution: λ = nf

!);(

k

ekf

k

fnfisol enfp );0(

* → “Optimal Transmission Ranges for Randomly Distributed Packet Radio Terminals”, by Takagi e Kleinrock (March 1984)


Connectivity analysis

hisolconn pp 1

*

apprd

Dh

dpd

dpd

l

D

disol

D

r

disol

appr

appr

2

12

0

2

12

0

)1(

)1(


Capacity Analysis

Gn

n

n

n

n

nnnB

conn

connconn

2

1

11

3

12

ni → connected nodes at i-hops distance from the BS


Experimental Results: Capacity

Averaged on 50 runs

variable nl = 2500r = 250

Avg. error:17,53 %

Avg. stdev:22,85 %

Cumulative throughput

0

500

1000

1500

2000

2500

3000

3500

0 0,5 1 1,5 2 2,5 3 3,5 4 4,5

# forward nodes

[kbi

t/sec

] analysis

simulation

[kb

it/se

c]


Conclusions

A first model for connectivity and capacity of WMNs was developed and validated

Good accuracy Open Points:

Inter-cell interferenceHeterogeneous WMN (e.g. WiMAX+WiFi)Node movement (statistical model)


TCP over Infrastructure WiFi

LL-ACK

TCP Data

Mobile Node(MN)

Base Station(BS)

Channel Contention

Channel Contention

LL-ACKTCP ACK

WiFi (IEEE 802.11)IP Network

TCP DataPHY

HeadersLL

Headers

TCP ACKPHY

HeadersLL

Headers

Legend:

Transmitted at Basic rateTransmitted at Data rate

ApplicatioApplication Datan Data

AcknowledgementAcknowledgements at different s at different

layerslayers


ARQ Proxy - Approach

Idea: To substitute the transmission of TCP ACK packets with a short MAC layer request on the radio link for multilayer ARQ overhead reduction


ARQ Proxy - Approach

TCPData

PHY /LLHeaders

LL-ACK

Fixed Host(FH)

IP Network

ARQ Proxy MAC ARQ ClientMAC TCP

TCP Data

Mobile Node(MN)

Base Station(BS)

TCP

o Access TCP header

o Get IP addr, port, etc.

o Generate TCP ACK & store

o Compute TCP ACK identification index

GenerateTCP ACK

TCP ACKIndex

TCP Data

TCP ACK

TCP ACKIndex


ARQ Proxy - Benefits

Performance and System Capacity Increase

LL-ACK

TCP Data

Mobile Node(MN)

Base Station(BS)

Channel Contention

Channel Contention

LL-ACKTCP ACK

WiFi (IEEE 802.11)IP Network

TCP DataPHYHeaders

LLHeaders

TCP ACKPHYHeaders

LLHeaders

Overhead Overhead reductionreduction


Evaluation Results

Throughput performance

0 300 600 900 1200 15000

1

2

3

4

5

6

TCP/IP datagram size (Bytes)

Th

rou

gh

pu

t (M

b/s

)

ARQ proxy ONARQ proxy OFF

Improvement: 20%

Improvement:up to 100%

VoIP and MultimediaApplications

TCP file transfer, Ethernet MTU


Evaluation Results

Round Trip Time (RTT) improvement

0 10 20 30 40 500

10

20

30

40

50

60

Simulation time (seconds)

Sm

oo

th R

TT

(m

sec)

ARQ proxy OFF

ARQ proxy ON

Order of millisecond

s


Evaluation Results

High error rate tolerance

0 0.05 0.1 0.15 0.2 0.25 0.3 0.350

1

2

3

4

5

6

7

Packet Error Rate (PER)

Th

rou

gh

pu

t (M

b/s

)

ARQ proxy ONARQ proxy OFF



UNITN:Testbed Activities

“WOMEN” Project Final MeetingTrento, Feb. 12th, 2008

RoofNet - Main Features: Wireless Mesh Networking SW developed by MIT

(based on Click Modular Router architecture) Multi-hop transmission Optimized routing using a specific metric

Limitations: Need for WiFi cards with Atheros chipset Madwifi driver

Goal: To port RoofNet to any hw/sw platform

Testbed Development


RoofNet - Why dedicated HW? Probe transmission for link quality measurements in

broadcast mode at arbitrary rate BSSID partitioning Routing metric is calculated by ETT

ETT = ETX * S/B

(S=data size, B=used bandwidth)

Problem: IEEE 802.11 does not enable arbitrary rate in

broadcast signalling

Testbed Development


Change from monitor mode to ad-hoc mode (supported by any IEEE 802.11 NIC)

Introduction of an element for static setting of preamble parameters

link-quality evaluation using ETX

flags = 0;

power = 60;

rssi = 8;

rate = 2;

retries = 0;

“BlankDecap” module settings

Roofnet Modifications


Service differentiation in Wi-Fi based Wireless Mesh NetworksSystem architectureSystem architecture Layer 2.5: interposition layer

with own headers ToS byte-based IP packet

classification Four classes with dedicated

buffering Transmissions scheduled by

a modifiedmodified Weighted Weighted Round Robin schemeRound Robin scheme

Software architectureSoftware architecture Click modular router Roofnet WMN features Elements added as C++

code

from Network Layer

to MAC layer

Layer 2.5

Routing probes


Experimental validation: results

1 hop1 hop 2 hops2 hops

UDP noise (diff)

UDP noise (no diff)

TCP hi-priority (no diff)

TCP hi-priority (diff)

Data-set C hi-priority TCP flow in class 4 UDP noise in class 1 at increasing bitrates

Figure 1: TCP resistance to UDP noise comparison (1 hop)

Figure 2: TCP resistance to UDP noise comparison (2 hops)


Experimental validation: results

Issues:Issues:1. Traffic generator issue:

TCP ACKS with ToS not set differentiation only on the feed-forward and

not on the feed-back ACKS scheduled with the lowest priority

2. Differentiation at Layer 2.5: Hi priority flows may lose MAC contentions

with low priority flows


WOMEN Testbed Implementation


AP3

AP4AP6

AP5

AP2

AP1

Control

Data



AP5

AP4

AP2

AP1

AP3

X

XX

XX



AP6X




UNITN:Dissemination


Dissemination activities - Project 11th Intenational Workshop on Computer-Aided

Modeling, Analysis and Design of Communication Links and Networks (CAMAD 2006, Trento, June 8-9, 2006)WOMEN leaflet CAMAD’06 website, dinner ticket Contribution to organization


Dissemination activities - Papers D. Kliazovich, F. Granelli, G. Pau, and M. Gerla

"APOHN: Subnetwork Layering to Improve TCP Performance over Heterogeneous Paths,“IEEE Next Generation Internet Design and Engineering (NGI), Valencia, Spain, April 2006.

D. Kliazovich, F. Granelli, and D. Miorandi"TCP Westwood+ Enhancements for High-Speed Long-Distance Networks,“IEEE International Conference on Communications (ICC'06), Istanbul, Turkey, June 2006.

F. Granelli, M. Devetsikiotis“Designing Cross-Layering Solutions for Wireless Networks: a General Framework and Its Application to a Voice-over-WiFi Scenario,”CAMAD 2006, June 8-9, 2006, pp. 1-7.

C.E. Costa, F. Granelli,"Optimal power control for embedded packet video transmission over WiFi“,1st Workshop on multiMedia Applications over Wireless Networks (MediaWIN 2006), Athens, Greece, April 2nd, 2006.


Dissemination activities - Papers E. Miorando, and F. Granelli

"On Connectivity and Capacity of Wireless Mesh Networks,“IEEE International Conference on Communications (ICC’07), Glasgow, Scotland.

D. Kliazovich, N. Ben Halima, F. Granelli,"Cross-layer error recovery optimization in WiFi networks“,2007 Tyrrhenian International Workshop on Digital Communications (TIWDC 2007), Island of Ischia, Italy, Sept. 9-12, 2007.

R. Riggio, N. Scalabrino, D. Miorandi, F. Granelli, Y. Fang, E. Gregori, I. Chlamtac,"Hardware and Software Solutions for Wireless Mesh Networks Testbeds“submitted to IEEE Communications Magazine (minor revision requested).

D. Kliazovich, F. Granelli, D. Miorandi,"Logarithmic Window Increase for TCP Westwood+ for Improvement in High Speed, Long Distance Networks“,submitted to Computer Networks Journal (minor revision requested).

“women” project final meeting trento, feb. 12th, 2008 1 unitn: description of work fabrizio...

Documents

total capacity slide

slide goal

performance optimization

tcp performance enhancement

cross layer interactions

rtt tcp logwestwood

smart tcp westwood window

wireless networks