0
Peering, QoS, and Price and
Quality Differentiation
J. Scott Marcus, Director
IDATE, Montpellier, 16 November 2011
1
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, transit, and Internet access (2)
ISP A ISP B ISP CISP 3
ISP C2 ISP C1
ISP B1
ISP C1a ISP A2
ISP A1b
ISP A2 ISP A1
ISP A1a ISP C1b
Peering Peering
ISP A ISP B ISP CISP 3
ISP C2 ISP C1
ISP B1
ISP C1a ISP A2
ISP A1b
ISP 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, 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
Stringent Latency Requirements
low high
highlow
sports, racing, shooter
strategy cards,
board games
role-playing
Bandwidth Requirements
Stringent Latency Requirements
low high
highlow
sports, racing, shooter
strategy cards,
board games
role-playing
Bandwidth Requirements
Stringent Latency Requirements
low high
highlow Email FTP
Video Streaming Audio
Streaming IRC
VoIP Video
Telephony
Channel Surfing
Browsing, Shopping, Banking
Bandwidth Requirements
Stringent Latency Requirements
low high
highlow Email FTP
Video Streaming Audio
Streaming IRC
VoIP Video
Telephony
Channel Surfing
Browsing, Shopping, Banking
Bandwidth Requirements
Stringent Latency Requirements
low high
highlow
sports, racing, shooter
strategy cards,
board games
role-playing
Bandwidth Requirements
Stringent Latency Requirements
low high
highlow
sports, racing, shooter
strategy cards,
board games
role-playing
Bandwidth Requirements
Stringent Latency Requirements
low high
highlow Email FTP
Video Streaming Audio
Streaming IRC
VoIP Video
Telephony
Channel Surfing
Browsing, Shopping, Banking
Bandwidth Requirements
Stringent Latency Requirements
low high
highlow Email FTP
Video Streaming Audio
Streaming IRC
VoIP Video
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 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)
Wait Time (microseconds) 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)
Wait Time (microseconds) 0.00
0.50 1.00 1.10 1.20 1.50 2.00 Coefficient of Variation
9
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
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
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 Google
Commercial ISP
Broadband ISP
User Google
Yahoo Yahoo
Bing Bing Google Google
Commercial ISP
Broadband 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 Google
Commercial ISP
Broadband ISP
User Google
Yahoo Yahoo
Bing Bing Google Google
Commercial ISP
Broadband 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
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%
3% 7%
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
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*
Data traffic/month (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*
Data traffic/month (PB)
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
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
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