Prof. Dr.-Ing. Dr. h. c. mult. P. J. Kühn Modeling and ... · Backoff Algorithm (2) Institute of Communication Networks and Computer Engineering University of Stuttgart • Modeled

INSTITUT FÜRNACHRICHTENVERMITTLUNG

UND DATENVERARBEITUNGProf. Dr.-Ing. Dr. h. c. mult. P. J. Kühn

Universität StuttgartINSTITUT FÜR

KOMMUNIKATIONSNETZEUND RECHNERSYSTEME

Prof. Dr.-Ing. Dr. h. c. mult. P. J. Kühn

Universität Stuttgart

Modeling and Performance Evaluationof a Manual Logon System

for Electronic Fee Collection

VDE/ITG-Workshop: Communication Applications for Logistics: Maut, Telematics & More

VDE/ITG FA 5.2., Bremen, January 26th., 2006

Joerg Sommer, Hanna Zündel, Christoph GaugerUniversity of Stuttgart, Institut für Kommunikationsnetze und Rechnersysteme

[email protected]

Steffen TackeDaimlerChrysler AG, Research and Technology

[email protected]

Institute of Communication Networks and Computer Engineering University of Stuttgart

Contents

❑ TollCollect’s logon processand system architecture

❑ Modeling aspects

❑ Performance evaluation

❑ Backoff Algorithm


• 2005, Germany introduced an Electronic Fee Collection System (EFC)

• Global navigation satellite system and cellular network (GNSS/CN)

• Currently > 700,000 registered usersBy 2012, might grow to over 9 million in Europe

• Three approaches for payment

Automatic Logon Manual Logon

• > 460,000(July 2005)

• approx. 3600 terminals

• approx. 80%revenue

• marginal anddeclining

• approx. 400 frequently usedterminals

• approx. 400 rarely usedterminals

On-boardunit

Internet andCall Centre

Toll-StationTerminal

German Toll Collection System "TollCollect"


TollCollect’s Manual Logon Process for EFC


• Billing data records have to be transferred quickly for enforcement

➥ System behavior after failure or outage is critical

TollCollect’s Manual Logon Process for EFC


Challenging Scenarios• System failure or breakdown of key components

• Billing and Data Centre

• Remote Access Server (RAS)

• Overload situation• Breakdown of the automatic GNSS/CN EFC system

• Specific and unexpected peaks

➥ Financial losses and negative standing for the operator

Aim of this work• Model and evaluate the overall manual logon process

regarding performance and scalability

• Optimize the algorithm and parameters for transmissionof billing data records after system outage

Manual Logon Process


Terminal-based EFC system architecture


Failure-free billing data records (BDR) delivering

TServerTimeout35 s.

TST RAS BDMUser/Driver

Timer abort

Tfollowup45 s.

Top

Connectionrequest

ACK

BDR

BDR

ACK

ACK

Connectiontermination

Tconnect~10 s.

Normal operational mode


Erroneous connection establishment

TServerTimeout35 s.

TST RAS BDMUser/Driver

Top

Connectionrequest

ACK

BDR

BDR


Tconnect~10 s.

Enterautonomous

mode

Autonomous mode


Billing data records (BDR) delivering

...

Parameters

• AutoReconnectIntervalARI

• AutoSendIntervalASI

• Number of deliveredBDRs n

• Time to transfer oneBDR tBDR

Relations

n

1 ASI t BDR<

ASIt BDR----------- ASI t BDR≥

⎩⎪⎨⎪⎧

=

RAS BDM

Connectionrequest

BDR

ACK


ACK

BDR

ACK

BDR

ACK

BDR

ACK

... ...

ARI

1 BDR

n BDR

Tfollowup45 s.

TST

ARI

Autonomous mode


Open queueing network

• Frontend comprises the terminals and the RAS

• Backend represents the BDM as a M/D/k-Multi-Server-Delay-System

THport...

FIFO...

FIFO

notification

Toll-StationTerminals

λIRS

λ1

λ2

λm

BDMRAS

IRS

Top

λBDM

λRASrp2

rps

rp1

...

D

D

D

...

M/D/kMulti-Server-Delay-System

2

k

1

Modeling the terminal-based EFC system


Scenario

Term

inal

que

ues

t

BDR arrivals

Normaloperation mode Autonomous mode

1 h outage Trecovery

Performance evaluation


Scenario

...

λIRS = 160 s-1

Class 1

λBDM...

...

Class 2

Class 3

Class 4

1920 RAS portsconnection requests 24 s-1

3600 terminals

k = 50h = 2 s

Term

inal

que

ues

t

BDR arrivals

Autonomous mode

1 h outage TrecoveryNormaloperation mode



Scenario

...

λIRS = 160 s-1

λBDM...

...

HomogeneousscenarioIAT = 180 s

1920 RAS portsconnection requests 24 s-1

3600 terminals

k = 50h = 2 s

Term

inal

que

ues

t

BDR arrivals

Autonomous mode

1 h outage TrecoveryNormaloperation mode



CDF of the queue processing (n = 1)

• With default configuration (ARI = 150 s) approximately Trecovery = 2.5 h

3000 4000 5000 6000 7000 8000 90000

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

t [s]

Dis

trib

utio

n of

term

inal

s do

ne

ARI = 150(default)



CDF of the queue processing (n = 1)

➥ Decreasing ARI values reduce Trecovery

3000 4000 5000 6000 7000 8000 90000

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

t [s]

Dis

trib

utio

n of

term

inal

s do

ne

ARI = 150(default)

ARI decreases

ARI = 100

ARI = 80

ARI = 60



BDM utilization (n = 1)

• ARI = 150 s can not utilize the BDM continuously

0 1000 2000 3000 4000 5000 6000 7000 80000

5

10

15

20

25

t [s]

Req

uest

s [1

/s]

24 s−1 RAS limit

ARI =150



BDM utilization (n = 1)

➥ Smaller ARI values are feasible due to RAS boundary

0 1000 2000 3000 4000 5000 6000 7000 80000

5

10

15

20

25

t [s]

Req

uest

s [1

/s]

24 s−1 RAS limit

ARI = 150

ARI = 100

ARI = 80

ARI = 60



So far, deterministic approach• ARI and n are constant values for all terminals

• Periodic system behavior → possible instability

• Default of the recovery algorithm is too restrictive

➥ The parameters after an outage are not optimal

➥ Challenge: improvement and optimization of the algorithm and parameters

➥ Aim: minimization of the recovery duration with controlled BDM load

Backoff Algorithm (1)


New approach for the backoff algorithm

Initial configuration in the autonomous mode ARI0=150 s, n0=2

Failure-free case Connection to RAS could be established successfully

Failure case Connection to RAS could not be established

ARIi+1 ARI0ARIi

2-----------+= n i+1

n i 2⋅ if 2n q l≤

q l else⎩⎨⎧

=

ARIi+1

ARIiARIi

2-----------– if ARIi

ARIi2

-----------–⎝ ⎠⎛ ⎞ ARI0≥

ARI0 else⎩⎪⎨⎪⎧

=

n i+1

n i

2---- if

ni

2---- n0≥

n0 else⎩⎪⎨⎪⎧

=

Backoff Algorithm (2)


• Modeled the manual logon process (users and system)

• Evaluated the recovery process in the autonomous mode

• Introduced a new approach for the backoff resolution

➥ Default manual logon process works stable,but is restrictive after an outage

• System behaviour depending on different downtime scenarios

• Optimization of the Backoff Algorithm to minimize RAS and BDMutilization

• Make the Backoff Algorithm dependant on state of

- terminal queue

- BDM

• Evaluate heterogeneous scenarios

Conclusion and outlook

INSTITUT FÜRNACHRICHTENVERMITTLUNG

UND DATENVERARBEITUNGProf. Dr.-Ing. Dr. h. c. mult. P. J. Kühn

Universität StuttgartINSTITUT FÜR

KOMMUNIKATIONSNETZEUND RECHNERSYSTEME

Prof. Dr.-Ing. Dr. h. c. mult. P. J. Kühn

Universität Stuttgart

Modeling and Performance Evaluationof a Manual Logon System

for Electronic Fee Collection

VDE/ITG-Workshop: Communication Applications for Logistics: Maut, Telematics & More

VDE/ITG FA 5.2., Bremen, January 26th., 2006

Joerg Sommer, Hanna Zündel, Christoph GaugerUniversity of Stuttgart, Institut für Kommunikationsnetze und Rechnersysteme

[email protected]

Steffen TackeDaimlerChrysler AG, Research and Technology

[email protected]

Documents

Prof. Dr.-Ing. Dr. h. c. mult. P. J. Kühn Modeling and ... · Backoff Algorithm (2) Institute of Communication Networks and Computer Engineering University of Stuttgart • Modeled