Peering, QoS, and Price and Quality Differentiation · 2012. 4. 11. · Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011 Traffic, costs, prices,

0

Peering, QoS, and

Price and

Quality Differentiation

J. Scott Marcus, Director

IDATE, Montpellier, 16 November 2011

1

Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011

Network Operators, Content Providers,

and the Open Internet

• Peering, transit, Internet access

• Quality of Service (QoS)

• A two-sided market view

• Traffic, costs, prices, and profitability

• Concluding observations

2


Peering, transit, and Internet access (1)

• Transit

- The customer pays the transit provider to provide connectivity

to substantially all of the Internet.

- Essentially the same service is provided to consumers,

enterprises, ISPs, content provider or application service

providers.

• Peering

- Two ISPs exchange traffic of their customers (and customers

of their customers).

- Often, but not always, done without charge.

• Variants of both exist.

3



ISP A ISP B ISP 3ISP C

ISP C2

ISP C1

ISP B1

ISP C1a

ISP A2

ISP A1bISP A2

ISP A1

ISP A1a ISP C1b

Peering PeeringISP A ISP B ISP 3ISP C

ISP C2

ISP C1

ISP B1

ISP C1a

ISP A2

ISP A1bISP A2

ISP A1

ISP A1a ISP C1b

Peering Peering

4



• Meanwhile, unit prices for global transit are declining rapidly.

• This decline reflects not only equipment costs but also circuits

(over land and under water).

• Labour and other OPEX elements play only a small role, since they

depend mostly on the number of subscribers.

Source: Telegeography (2011), WIK calculations.

5



Shortest exit routing

Copyright (c) 1999 by Addison Wesley Longman, Inc. All rights reserved. No part of this publication

may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic,

mechanical, photocopying, recording, or otherwise, without the prior consent of the publisher.

New York

Chicago

Atlanta

Dallas

Los Angeles

San Francisco Washington DC

Web Site

Peering Point

Peering Point

Backbone ISP

Backbone ISP

6


Interconnection (QoS)

• Real time bidirectional audio: stringent requirements

• Email: liberal requirements

• Streamed audio and video: fairly liberal requirements.

(Channel surfing?)

Bandwidth Requirements

Str

ingen

t L

ate

ncy R

equ

irem

en

ts

low high

hig

hlo

w

sports, racing, shooter

strategy

cards,

board games

role-playing


Str

ingen

t L

ate

ncy R

equ

irem

en

ts

low high

hig

hlo

w


strategy

cards,

board games

role-playing


Str

inge

nt

La

ten

cy R

eq

uir

em

en

ts

low high

hig

hlo

w Email FTP

Video

Streaming

Audio

StreamingIRC

VoIPVideo

Telephony

Channel Surfing

Browsing,

Shopping,

Banking


Str

inge

nt

La

ten

cy R

eq

uir

em

en

ts

low high

hig

hlo

w Email FTP

Video

Streaming

Audio

StreamingIRC

VoIPVideo

Telephony

Channel Surfing

Browsing,

Shopping,

Banking


Str

ingen

t L

ate

ncy R

equ

irem

en

ts

low high

hig

hlo

w


strategy

cards,

board games

role-playing


Str

ingen

t L

ate

ncy R

equ

irem

en

ts

low high

hig

hlo

w


strategy

cards,

board games

role-playing


Str

inge

nt

La

ten

cy R

eq

uir

em

en

ts

low high

hig

hlo

w Email FTP

Video

Streaming

Audio

StreamingIRC

VoIPVideo

Telephony

Channel Surfing

Browsing,

Shopping,

Banking


Str

inge

nt

La

ten

cy R

eq

uir

em

en

ts

low high

hig

hlo

w Email FTP

Video

Streaming

Audio

StreamingIRC

VoIPVideo

Telephony

Channel Surfing

Browsing,

Shopping,

Banking

General Applications Gaming Applications

7


Shifts in Internet traffic

Source: Cisco (2011).

• Voice drives revenue, but is a declining fraction of traffic.

• Concerns have been voiced in recent years over the explosion of

video traffic over the Internet, and its implications for network cost.

8



M/G/1 queueing analysis of the performance of a single link

(with clocking delay of 50 μsecs (284 byte packets) and a 155 Mbps link)

M/G/1 Queuing Delay (155 Mbps Link)

-

50.00

100.00

150.00

200.00

250.00

300.00

350.00

0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90

Utilization (rho)

Wa

it T

ime

(m

icro

se

co

nd

s)

0.00

0.50

1.00

1.10

1.20

1.50

2.00

Coefficient

of Variation

M/G/1 queueing analysis of the performance of a single link (with clocking delay of 50 μsecs (284 byte packets) and a 155 Mbps link)

M/G/1 Queuing Delay (155 Mbps Link)

-

50.00

100.00

150.00

200.00

250.00

300.00

350.00

0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90

Utilization (rho)

Wa

it T

ime

(m

icro

se

co

nd

s)

0.00

0.50

1.00

1.10

1.20

1.50

2.00

Coefficient

of Variation

9



• As we have seen, per-hop delay, even in a network with 90% load,

is about 1,000 times less than the 150 millisecond delay “budget”

for real-time bidirectional voice.

• IMPLICATION: Most of the time, and under normal conditions,

variable delay in the core of the network is unlikely to be perceptible

to the users of VoIP or other delay-sensitive applications.

• FURTHER IMPLICATION: Consumers will not willingly pay a large

premium for a performance difference that they cannot perceive.

• Packet delay is more likely to be an issue:

- For slower circuits at the edge of the network

- For shared circuits (e.g. cable modem services)

- When one or more circuits are saturated

- When one or more components have failed

- When a force majeure incident has occurred

10


Implementing inter-provider QoS

• Although the technology is reasonably straightforward,

little practical experience in enforcing QoS across IP-

based networks.

• It is not due to a lack of standards – there are too many

standards, not too few.

• Classic problem of introducing change into a

technological environment:

- Network effects – no value until enough of the market has

switched.

- Long, complex value chains.

- Costs and complexity of transition.

• Analogous problems have slowed IPv6 and DNSSEC.

11



• Efforts to extend Quality of Service (QoS) across network

operators have failed to catch fire for many reasons:

- Scale: Bilateral peering arrangements will tend to be acceptable to

both network operators only when the networks are of similar scale,

or more precisely when both networks can be expected to be

subject to similar cost drivers for carrying their respective traffic.

- Traffic balance: Where traffic is significantly asymmetric, cost

drivers are likely to also be asymmetric.

- Monitoring and management: There are many practical

challenges in determining whether each network operator has in

fact delivered the QoS that it committed to deliver.

- Financial arrangements: There has been no agreement as to how

financial arrangements should work. In particular, there has been

enormous reluctance on the part of network operators to accept

financial penalties for failing to meet quality standards.

12



• Many efforts over the years to define inter-provider QoS

standards.

• One of the best and most practical was organised by MIT, with

substantial industry participation.

• The following values from the MIT white paper would appear to

be resonable for IP interconnection suitable for real time

bidirectional voice:

Delay: 100 msec

Delay Variance: 50 msec

IPPM Loss Ratio: 1 x 10-3 (One Way Packet Loss)

• The MIT WG white paper also explains how to measure these,

and how to allocate end-to-end requirements to multiple

networks. IPPM probes could be suitable.

• A challenge: No network operator will want another to operate

probes within its network.

13



• As part of the functional/operational separation of Telecom New

Zealand, there were commitments

- To interconnect with competitors using IP

- To support a suitable QoS for VoIP in those interconnections

• The first of these is in place.

• For the second, Telecom New Zealand made a quite interesting

proposal, based on their methodology for the first of these.

14

Quality differentiation and network neutrality

• Quality differentiation

• Economic foreclosure

• Two-sided (or multi-sided) markets

15

Quality differentiation

• Quality differentiation and price differentiation are well understood

practices.

• In the absence of anticompetitive discrimination, differentiation

generally benefits both producers and consumers.

• We typically do not consider it problematic if an airline or rail service

offers us a choice between first class and second class seats.

16

Economic foreclosure

• When a producer with market

power in one market segment

attempts to project that market

power into upstream or

downstream segments that

would otherwise be competitive,

that constitutes economic

foreclosure.

• Foreclosure harms consumers,

and imposes an overall socio-

economic deadweight loss on

society.

Google

Yahoo

Bing

Google

Commercial

ISPBroadband

ISP

User

Google

YahooYahoo

BingBing

GoogleGoogle

Commercial

ISPBroadband

ISP

User

17

Two-sided markets

• The Internet can be thought of

as a two-sided market, with

network operators serving as a

platform connecting providers of

content (e.g. web sites) with

consumers.

• Under this view, some disputes

are simply about how costs and

profits should be divided

between the network operators

and the two (or more) sides of

the market.

Google

Yahoo

Bing

Google

Commercial

ISPBroadband

ISP

User

Google

YahooYahoo

BingBing

GoogleGoogle

Commercial

ISPBroadband

ISP

User

18


Traffic, costs, prices, and profitability (1)

• A.T. Kearney (2010):

“Internet traffic is exploding in an unprecedented way due to

increasing use of video. Costs for network operators are sky-

rocketing, even under existing technology and even without

considering the huge investments needed for fibre-based Next

Generation Access. Due to market defects, there is no way to

make consumers shoulder the cost of the increased bandwidth;

thus, it will soon become necessary for firms that provide content

to pay for the network for the first time, much as content and

advertising typically pay for over-the-air broadcast television.”

• Intuitive? Satisfying? Plausible?

19



• Traffic growth is largely a function of:

- an increase in the number of subscribers, and

- an increase in traffic per subscriber.

• Some costs are largely driven by the number of subscribers, and

are largely independent of usage per subscriber.

• Unit costs for network equipment in the core and concentration

networks (including routers and optoelectronics), where costs are

usage-dependent, are declining at a rate comparable to that of

Internet traffic increase per user in the fixed network. This can be

viewed as an example of Moore’s Law.

• Cost per customer and revenue per customer in the fixed network

remain in balance, despite the increase in traffic.

20



• The core network is about 7% of total cost, the concentration

network about 6%.

• Both benefit from these technological enhancements.

Source: German BNetzA (2009).

6%

31%

14%

3%

6%

7% 3%

9%

21%

Monthly cost of a bundled DSL broadband/voice service (BNetzA 2009)

Wholesale price ULL Provision

Monthly rental ULL

Cost of DSLAM

Cost of splitter

Transport in concentration network

Transport IP backbone network

Collocation

Usage dependent cost telephony

Customer acquisition, maintenance, common cost,billing, bad debt

21



Fixed Data Traffic

0

5000

10000

15000

20000

25000

30000

35000

40000

2006 2007 2008 2009 2010 2011* 2012*

Dat

a tr

affi

c/m

on

th (

PB

)

Middle East and Africa

Latin America

Japan

Asia Pacific

North America

Central Eastern Europe

Western Europe

Mobile Data Traffic

0

200

400

600

800

1000

1200

1400

2006 2007 2008 2009 2010 2011* 2012*

Da

ta t

raff

ic/

mo

nth

(P

B)

Middle East and Africa

Latin America

Japan

Asia Pacific

North America

Central Eastern Europe

Western Europe

Traffic is indeed increasing in both

the fixed and the mobile networks.

Source: Cisco (2011), WIK calculations.

22



• However, the rate of growth in percentage terms is declining over

time.

Source: Cisco (2011), WIK calculations.

23



0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

18,000

20,000

2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015

$0.00

$5.00

$10.00

$15.00

$20.00

$25.00

Worldwide high end router capacity shipped (Mbps)

Price per Mbps (USD)

0

1,000

2,000

3,000

4,000

5,000

6,000

7,000

2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015

$0.00

$1.00

$2.00

$3.00

$4.00

$5.00

$6.00

$7.00

Worldwide long haul DWDM capacity shipped (Mbps)Price per Mbps (USD)

Here we have the shipment

quantities in Mbps and the price

per Mbps (USD) for high end

routers and for long haul DWDM

optoelectronic equipment.

These are among the key cost

drivers for Internet core and

aggregation networks.

The growth in shipments generally

tracks the Cisco projections.

The growth in shipment volume

does not equate to a growth in

costs, because the decline in unit

costs is nearly in balance with it.

Source: Dell’Oro (2011), WIK calculations.

24



• The trend in underlying equipment costs (and many other costs)

tracks subscribership and revenue, not with the volume of traffic.

Source: Dell‘Oro (2011), Cisco (2011), WIK calculations.

-

200

400

600

800

1,000

1,200

1,400

1,600

1,800

2,000

2006 2007 2008 2009 2010 2011 2012 2013 2014 2015

IP traffic Fixed BB subscribers

SP Edge total revenue SP Core total revenue

DWDM Long Haul total revenue WDM Metro total revenue

25


Concluding observations

• In competitive markets, quality differentiation typically benefits both

suppliers and producers.

• Beware quick fixes! If a solution seems too good to be true, it

probably is.

• Market mechanisms often reach better solutions than well

intentioned policymakers.

wik-Consult GmbH

Postfach 2000

53588 Bad Honnef

Tel 02224-9225-0

Fax 02224-9225-68

eMail [email protected]

www.wik-consult.com

Documents

Peering, QoS, and Price and Quality Differentiation · 2012. 4. 11. · Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011 Traffic, costs, prices,