quality of service in telecommunícation networks… · quality of service in telecommunication...

36
Quality of Service in Telecommunication Networks Åke Arvidsson, Ph.D. Ericsson Core Network Development, Sweden

Upload: dinhbao

Post on 29-Mar-2018

213 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: Quality of Service in Telecommunícation Networks… · Quality of Service in Telecommunication Networks ... Can be used to control quality of service. ... Earlier diagram based on

Quality of Service inTelecommunication Networks

Åke Arvidsson, Ph.D.Ericsson Core Network Development, Sweden

Page 2: Quality of Service in Telecommunícation Networks… · Quality of Service in Telecommunication Networks ... Can be used to control quality of service. ... Earlier diagram based on

© Ericsson AB 2005 ÄS/EAB/UKT/T Åke Arvidsson Quality of Service in Telecommunícation Networks 2005-11-082

Main Message

Traffic theory:– QoS is (partly) about congestion.– Congestion is partly caused by temporary overload.– Temporary overloads can be handled by statistics.– QoS can be controlled by statistics.

Typical issues for regulators:– Is an operator serious about quality of service?– What is the fair price to carry additional traffic?

Typical issues for operators:– How should a network be engineered?– Can I just go for “over-provisioning”!

Page 3: Quality of Service in Telecommunícation Networks… · Quality of Service in Telecommunication Networks ... Can be used to control quality of service. ... Earlier diagram based on

© Ericsson AB 2005 ÄS/EAB/UKT/T Åke Arvidsson Quality of Service in Telecommunícation Networks 2005-11-083

Overview

Background and motivation.The significance of variations.Big variations means expenses.

– Relationship to shared networks.

Small variations means savings.– Relationship to shared networks.

Hot topic: Relevance to IP and integrated services.

Page 4: Quality of Service in Telecommunícation Networks… · Quality of Service in Telecommunication Networks ... Can be used to control quality of service. ... Earlier diagram based on

© Ericsson AB 2005 ÄS/EAB/UKT/T Åke Arvidsson Quality of Service in Telecommunícation Networks 2005-11-084

Background (1)

Voice and data traffic exhibit significant variations.Variable demand but fixed resources.

– Traffic may be lost.– Traffic may be delayed.– Traffic may be subject to other impairments.

Mathematically tractable by traffic theory.

Page 5: Quality of Service in Telecommunícation Networks… · Quality of Service in Telecommunication Networks ... Can be used to control quality of service. ... Earlier diagram based on

© Ericsson AB 2005 ÄS/EAB/UKT/T Åke Arvidsson Quality of Service in Telecommunícation Networks 2005-11-085

Background (2)

Can be used to control quality of service.Essential part of competition.

– Competitive advantage.– Add value to services.

Essential part of regulation/competition.– Fair treatment of new players by former monopolies.– Fair criteria for customers to choose service provider.

Page 6: Quality of Service in Telecommunícation Networks… · Quality of Service in Telecommunication Networks ... Can be used to control quality of service. ... Earlier diagram based on

© Ericsson AB 2005 ÄS/EAB/UKT/T Åke Arvidsson Quality of Service in Telecommunícation Networks 2005-11-086

Background (3)

Significance:– Society: Reliable services.– Operators: Stable systems, fair competition.– Users: Value for money.

Page 7: Quality of Service in Telecommunícation Networks… · Quality of Service in Telecommunication Networks ... Can be used to control quality of service. ... Earlier diagram based on

© Ericsson AB 2005 ÄS/EAB/UKT/T Åke Arvidsson Quality of Service in Telecommunícation Networks 2005-11-087

Background (4)

Requirements:– Prerequisite: The busy hour concept.– Delays: dialing tone, through connection, speech, ...

ITU-T, ETSI and others.– Blocking: lost traffic, ...

Typically operator dependent.– The above apply to voice services.

Data services specified but less used.

Page 8: Quality of Service in Telecommunícation Networks… · Quality of Service in Telecommunication Networks ... Can be used to control quality of service. ... Earlier diagram based on

© Ericsson AB 2005 ÄS/EAB/UKT/T Åke Arvidsson Quality of Service in Telecommunícation Networks 2005-11-088

Variations

The “problem” lies in variations:– Ideal: Fixed inter-arrival times and fixed service times:

Time

– Real: Variable inter-arrival times (and variable service times):

Time

Page 9: Quality of Service in Telecommunícation Networks… · Quality of Service in Telecommunication Networks ... Can be used to control quality of service. ... Earlier diagram based on

© Ericsson AB 2005 ÄS/EAB/UKT/T Åke Arvidsson Quality of Service in Telecommunícation Networks 2005-11-089

Traffic Theory

Statistical analysis of service systems:– Waiting times.

Example: Dialing tone delay in a switch.– Congestion probabilities.

Example: Traffic rejected from a trunk.– Utilisations.

Example: Load on a processor or link.

Page 10: Quality of Service in Telecommunícation Networks… · Quality of Service in Telecommunication Networks ... Can be used to control quality of service. ... Earlier diagram based on

© Ericsson AB 2005 ÄS/EAB/UKT/T Åke Arvidsson Quality of Service in Telecommunícation Networks 2005-11-0810

Quality of service vs. supply

0,0

0,2

0,4

0,6

0,8

1,0

0,6 0,7 0,8 0,9 1,0 1,1 1,2 1,3 1,4

Capacity/Demand

Prob

abili

ty o

f suf

ficie

ncy

Better QoStakes more

capacityOnly0.53

No “over-supply”

Page 11: Quality of Service in Telecommunícation Networks… · Quality of Service in Telecommunication Networks ... Can be used to control quality of service. ... Earlier diagram based on

© Ericsson AB 2005 ÄS/EAB/UKT/T Åke Arvidsson Quality of Service in Telecommunícation Networks 2005-11-0811

Conclusions (1)

Dimension for average:– Variations unaccounted for.– Related impairments ignored.– No control of QoS!

Traffic theory necessary for QoS.– Correct provisioning aims at targeted QoS.– Over-provisioning is more than that, not just more than the

mean.

Page 12: Quality of Service in Telecommunícation Networks… · Quality of Service in Telecommunication Networks ... Can be used to control quality of service. ... Earlier diagram based on

© Ericsson AB 2005 ÄS/EAB/UKT/T Åke Arvidsson Quality of Service in Telecommunícation Networks 2005-11-0812

Randomness (1)

More randomness means more problems.– More delay, higher blocking, etc.

Examples:– Circuit oriented traffic.

Erlang: Number of parallel connections.– Packet oriented traffic.

Load: A metric between zero and unit.

Page 13: Quality of Service in Telecommunícation Networks… · Quality of Service in Telecommunication Networks ... Can be used to control quality of service. ... Earlier diagram based on

© Ericsson AB 2005 ÄS/EAB/UKT/T Åke Arvidsson Quality of Service in Telecommunícation Networks 2005-11-0813

Blocking and Variations

0.000

0.050

0.100

0.150

0.200

0.250

0.300

0.350

0.00 2.00 4.00 6.00 8.00 10.00

Traffic variation (peakedness)

Blo

ckin

g pr

obab

ility

1 erlang

10 erlang

100 erlang

Page 14: Quality of Service in Telecommunícation Networks… · Quality of Service in Telecommunication Networks ... Can be used to control quality of service. ... Earlier diagram based on

© Ericsson AB 2005 ÄS/EAB/UKT/T Åke Arvidsson Quality of Service in Telecommunícation Networks 2005-11-0814

Waiting Time and Variations

0

50

100

150

200

0 2 4 6 8 10

Service time variation

Mea

n w

aitin

g tim

e

Load 0.5

Load 0.8

Load 0.2

Page 15: Quality of Service in Telecommunícation Networks… · Quality of Service in Telecommunication Networks ... Can be used to control quality of service. ... Earlier diagram based on

© Ericsson AB 2005 ÄS/EAB/UKT/T Åke Arvidsson Quality of Service in Telecommunícation Networks 2005-11-0815

Conclusions (2)

More randomness means more problems.– Split costs depend on traffic variations.

Simple “dimensioning” based on factors:– Correct answers for (at most) one working point only.

“Over-dimensioning” is not as simple as it may sound.– How much is too much?

Page 16: Quality of Service in Telecommunícation Networks… · Quality of Service in Telecommunication Networks ... Can be used to control quality of service. ... Earlier diagram based on

© Ericsson AB 2005 ÄS/EAB/UKT/T Åke Arvidsson Quality of Service in Telecommunícation Networks 2005-11-0816

Randomness (2)

Less randomness means less problems.– Less delay, lower blocking, etc.

Examples:– Circuit oriented traffic.

Erlang: Number of parallel connections.– Packet oriented traffic.

Load: A metric between zero and unit.

Page 17: Quality of Service in Telecommunícation Networks… · Quality of Service in Telecommunication Networks ... Can be used to control quality of service. ... Earlier diagram based on

© Ericsson AB 2005 ÄS/EAB/UKT/T Åke Arvidsson Quality of Service in Telecommunícation Networks 2005-11-0817

Circuits per Erlang

Circuit oriented traffic.– How many circuits are needed for one Erlang?– Depends on performance and the scale of the system!

0.50.0

1

1.01.52.02.53.03.54.04.55.0

10 100 1000 10000

Traffic

Circ

uits

per

erla

ng

High grade of service (0.5%)

Low grade of service (5%)

Page 18: Quality of Service in Telecommunícation Networks… · Quality of Service in Telecommunication Networks ... Can be used to control quality of service. ... Earlier diagram based on

© Ericsson AB 2005 ÄS/EAB/UKT/T Åke Arvidsson Quality of Service in Telecommunícation Networks 2005-11-0818

Packet Delay

Packet oriented traffic.– What is the delay for a packet/signal?– Depends on the load and the scale of the system!

0.00001

0.001

0.1

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Load

Del

ay (s

)

30 parallel 64 kbps links;random/optimal sharing(20 octet packets)One 30×64 kbps link(20 octet packets)

Page 19: Quality of Service in Telecommunícation Networks… · Quality of Service in Telecommunication Networks ... Can be used to control quality of service. ... Earlier diagram based on

© Ericsson AB 2005 ÄS/EAB/UKT/T Åke Arvidsson Quality of Service in Telecommunícation Networks 2005-11-0819

Conclusions (3)

Less randomness means less problems.– Sharing networks reduces marginal costs.

“The bigger the better”.– One big system is better than many small systems.

Law of large numbers.– The more samples, the less variation.

Page 20: Quality of Service in Telecommunícation Networks… · Quality of Service in Telecommunication Networks ... Can be used to control quality of service. ... Earlier diagram based on

© Ericsson AB 2005 ÄS/EAB/UKT/T Åke Arvidsson Quality of Service in Telecommunícation Networks 2005-11-0820

Classic Traffic Theory

Circuit switched services.– Erlang-B.

Agner Krarup Erlang(1878 – 1929)

KTAS, Copenhagen, 1917.

Pleased tomeet you!

– Wide range of extensions.

Packet switched services.– Signal networks.– Early arpanet community.

How about present IP?– Before: Best effort only.– Now: QoS required.

Page 21: Quality of Service in Telecommunícation Networks… · Quality of Service in Telecommunication Networks ... Can be used to control quality of service. ... Earlier diagram based on

© Ericsson AB 2005 ÄS/EAB/UKT/T Åke Arvidsson Quality of Service in Telecommunícation Networks 2005-11-0821

Quality of Service in IP

Profitability:– Charging on QoS to boost income.

Integration:– Service integration to cut expenses.

Efficiency:– Utilise expensive wireless resources.

Page 22: Quality of Service in Telecommunícation Networks… · Quality of Service in Telecommunication Networks ... Can be used to control quality of service. ... Earlier diagram based on

© Ericsson AB 2005 ÄS/EAB/UKT/T Åke Arvidsson Quality of Service in Telecommunícation Networks 2005-11-0822

Recent Developments (1)

Data (IP) traffic models:– Packets are like calls in telephone traffic.– Poisson process.– “Burstier” processes.

Ethernet (and other) measurements:– High spread (heavy tails).– Slow variations (long range dependence).– Time scale independence (self similarity).

Earlier models far too optimistic!– Forget the old economy, the new economy is here!

Page 23: Quality of Service in Telecommunícation Networks… · Quality of Service in Telecommunication Networks ... Can be used to control quality of service. ... Earlier diagram based on

© Ericsson AB 2005 ÄS/EAB/UKT/T Åke Arvidsson Quality of Service in Telecommunícation Networks 2005-11-0823

Results (1)

Heaviesttail

Lightesttail

Load≈71%

Queue10,000

Noqueue

Load≈ 50%

Page 24: Quality of Service in Telecommunícation Networks… · Quality of Service in Telecommunication Networks ... Can be used to control quality of service. ... Earlier diagram based on

© Ericsson AB 2005 ÄS/EAB/UKT/T Åke Arvidsson Quality of Service in Telecommunícation Networks 2005-11-0824

Recent Developments (2)

Objections:– Most data traffic is subject to flow control.

Queues cannot grow without bounds.– TCP applies dynamic flow control to maximise throughput.

Links may operate near (local) saturation.

Earlier diagram based on measured traffic.– Measurements are reality but– the experiments are not real since flow control is “frozen”.

Page 25: Quality of Service in Telecommunícation Networks… · Quality of Service in Telecommunication Networks ... Can be used to control quality of service. ... Earlier diagram based on

© Ericsson AB 2005 ÄS/EAB/UKT/T Åke Arvidsson Quality of Service in Telecommunícation Networks 2005-11-0825

Results (2)

Earlier diagram Active flow control

Page 26: Quality of Service in Telecommunícation Networks… · Quality of Service in Telecommunication Networks ... Can be used to control quality of service. ... Earlier diagram based on

© Ericsson AB 2005 ÄS/EAB/UKT/T Åke Arvidsson Quality of Service in Telecommunícation Networks 2005-11-0826

Conclusions (4)

“New” discoveries not so dramatic.However, traffic and systems interaction complicates:

– Measurements:What does link load mean to user performance?

– Modeling:Simple steady state Poisson not generally applicable.

– Dimensioning:Traffic and performance mutually dependent.

Page 27: Quality of Service in Telecommunícation Networks… · Quality of Service in Telecommunication Networks ... Can be used to control quality of service. ... Earlier diagram based on

© Ericsson AB 2005 ÄS/EAB/UKT/T Åke Arvidsson Quality of Service in Telecommunícation Networks 2005-11-0827

Recent Developments (3)

Research at Ericsson:– Mathematical methods.

Performance criterion:– Downloading time (useful throughput).

Examples:– Bottleneck identification and removal.– “Peak factor” calculations.

Page 28: Quality of Service in Telecommunícation Networks… · Quality of Service in Telecommunication Networks ... Can be used to control quality of service. ... Earlier diagram based on

© Ericsson AB 2005 ÄS/EAB/UKT/T Åke Arvidsson Quality of Service in Telecommunícation Networks 2005-11-0828

Example: Bottlenecks (1)

05

101520253035404550

Access ratePacket dropPropagationW

indow sizeFile size

+ %

10 packetsFile size:

4 packetsWindow size:

100 msPropagation:

2 %Packet drop:

256 kbpsAccess rate:

10 packetsFile size:

4 packetsWindow size:

50 msPropagation:

2 %Packet drop:

256 kbpsAccess rate:

10 packetsFile size:

4 packetsWindow size:

50 msPropagation:

2 %Packet drop:

256 kbpsAccess rate:

10 packetsFile size:

4 packetsWindow size:

50 msPropagation:

2 %Packet drop:

256 kbpsAccess rate:

10 packetsFile size:

4 packetsWindow size:

50 msPropagation:

2 %Packet drop:

256 kbpsAccess rate:

94 kbpsThroughput: 130 kbpsThroughput:

Page 29: Quality of Service in Telecommunícation Networks… · Quality of Service in Telecommunication Networks ... Can be used to control quality of service. ... Earlier diagram based on

© Ericsson AB 2005 ÄS/EAB/UKT/T Åke Arvidsson Quality of Service in Telecommunícation Networks 2005-11-0829

Example: Bottlenecks (2)

05

101520253035404550

Access ratePacket dropPropagationW

indow sizeFile size

10 packetsFile size:

4 packetsWindow size:

50 msPropagation:

2 %Packet drop:

256 kbpsAccess rate:

20 packetsFile size:

4 packetsWindow size:

50 msPropagation:

2 %Packet drop:

256 kbpsAccess rate:+ %

20 packetsFile size:

4 packetsWindow size:

50 msPropagation:

2 %Packet drop:

256 kbpsAccess rate:

20 packetsFile size:

4 packetsWindow size:

50 msPropagation:

2 %Packet drop:

256 kbpsAccess rate:

20 packetsFile size:

4 packetsWindow size:

50 msPropagation:

2 %Packet drop:

256 kbpsAccess rate:

05

101520253035404550

Access ratePacket dropPropagationW

indow sizeFile size

130 kbpsThroughput: 165 kbpsThroughput:

Page 30: Quality of Service in Telecommunícation Networks… · Quality of Service in Telecommunication Networks ... Can be used to control quality of service. ... Earlier diagram based on

© Ericsson AB 2005 ÄS/EAB/UKT/T Åke Arvidsson Quality of Service in Telecommunícation Networks 2005-11-0830

Example: Bottlenecks (3)

05

101520253035404550

Access ratePacket dropPropagationW

indow sizeFile size

20 packetsFile size:

4 packetsWindow size:

50 msPropagation:

2 %Packet drop:

256 kbpsAccess rate:

20 packetsFile size:

4 packetsWindow size:

50 msPropagation:

2 %Packet drop:

512 kbpsAccess rate:+ %

20 packetsFile size:

4 packetsWindow size:

50 msPropagation:

2 %Packet drop:

512 kbpsAccess rate:

20 packetsFile size:

4 packetsWindow size:

50 msPropagation:

2 %Packet drop:

512 kbpsAccess rate:

20 packetsFile size:

4 packetsWindow size:

50 msPropagation:

2 %Packet drop:

512 kbpsAccess rate:

05

101520253035404550

Access ratePacket dropPropagationW

indow sizeFile size

165 kbpsThroughput: 216 kbpsThroughput:

Page 31: Quality of Service in Telecommunícation Networks… · Quality of Service in Telecommunication Networks ... Can be used to control quality of service. ... Earlier diagram based on

© Ericsson AB 2005 ÄS/EAB/UKT/T Åke Arvidsson Quality of Service in Telecommunícation Networks 2005-11-0831

Example: “Peak Factor” (1)

Network:– Packet loss probability 1%.– Round trip time 300 ms.

Users:– Access rate: 33% 64/128 kbps; 67% 64/384 kbps.– Packet size: 25% 536 bytes; 75% 1540 bytes.– Window size: 16.384 kbyte.

Traffic:– 170 kbyte/busy hour.– WAP, WWW, MMS, Mail.

Page 32: Quality of Service in Telecommunícation Networks… · Quality of Service in Telecommunication Networks ... Can be used to control quality of service. ... Earlier diagram based on

© Ericsson AB 2005 ÄS/EAB/UKT/T Åke Arvidsson Quality of Service in Telecommunícation Networks 2005-11-0832

Example: “Peak Factor” (2)

Dimension core network:– A connection sees 5 links in tandem end-to-end.– 5% additional RTT acceptable (15 ms means 3 ms per link).– Different links have different load.

One peak factor?– Twice the average (peak factor 2, load 50%).– Is this too much (over-dimensioning)?– Or do we need more (e, π, ...)?

Page 33: Quality of Service in Telecommunícation Networks… · Quality of Service in Telecommunication Networks ... Can be used to control quality of service. ... Earlier diagram based on

© Ericsson AB 2005 ÄS/EAB/UKT/T Åke Arvidsson Quality of Service in Telecommunícation Networks 2005-11-0833

Example: “Peak Factor” (3)

0

2

4

6

8

10

12

14

16

18

0 2000 4000 6000 8000 10000

Sharing users

Peak

fact

or

Page 34: Quality of Service in Telecommunícation Networks… · Quality of Service in Telecommunication Networks ... Can be used to control quality of service. ... Earlier diagram based on

© Ericsson AB 2005 ÄS/EAB/UKT/T Åke Arvidsson Quality of Service in Telecommunícation Networks 2005-11-0834

Conclusions (5)

IP (TCP) fractal properties often overrated.– Relationship to variation conceptually the same.

This does not mean that dimensioning is simple:– Traffic and network are mutually dependent.– Traffic exhibits large variations over time.– User perceived performance different from directly

measurable.– A magnitude of parameters means unclear bottlenecks.– High burstiness may still enforce low utilisation.

Page 35: Quality of Service in Telecommunícation Networks… · Quality of Service in Telecommunication Networks ... Can be used to control quality of service. ... Earlier diagram based on

© Ericsson AB 2005 ÄS/EAB/UKT/T Åke Arvidsson Quality of Service in Telecommunícation Networks 2005-11-0835

Main Message

Traffic theory:– QoS is (partly) about congestion (loss and delay).– Congestion is partly caused by temporary overload

(variations).– Temporary overloads can be handled by statistics (traffic

model).– QoS can be controlled by statistics (queuing theory).

Typical issues:– Is an operator serious about quality of service? Goals!– What is the fair price to carry additional traffic? Depends!– How should a network be engineered? Methods!– Can I trust “over-provisioning”! No!

Page 36: Quality of Service in Telecommunícation Networks… · Quality of Service in Telecommunication Networks ... Can be used to control quality of service. ... Earlier diagram based on

© Ericsson AB 2005 ÄS/EAB/UKT/T Åke Arvidsson Quality of Service in Telecommunícation Networks 2005-11-0836