Post on 17-Oct-2020
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, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
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
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
Peering, transit, and Internet access (2)
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
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
Peering, transit, and Internet access (3)
• 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
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
Peering, transit, and Internet access (4)
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
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
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
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
Bandwidth Requirements
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
Bandwidth Requirements
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
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
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
Bandwidth Requirements
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
Bandwidth Requirements
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
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
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
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
Interconnection (QoS)
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
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
Interconnection (QoS)
• 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
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
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
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
Implementing inter-provider QoS
• 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
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
Implementing inter-provider QoS
• 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
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
Implementing inter-provider QoS
• 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.
Yahoo
Bing
Commercial
ISPBroadband
ISP
User
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.
Yahoo
Bing
Commercial
ISPBroadband
ISP
User
YahooYahoo
BingBing
GoogleGoogle
Commercial
ISPBroadband
ISP
User
18
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
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
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
Traffic, costs, prices, and profitability (2)
• 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
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
Traffic, costs, prices, and profitability (3)
• 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
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
Traffic, costs, prices, and profitability (4)
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
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
Traffic, costs, prices, and profitability (5)
• However, the rate of growth in percentage terms is declining over
time.
Source: Cisco (2011), WIK calculations.
23
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
Traffic, costs, prices, and profitability (6)
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
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
Traffic, costs, prices, and profitability (7)
• 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
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
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 info@wik-consult.com
www.wik-consult.com