Azure Monitor & Application Insight to monitor Infrastructure & Application
Mr Marcus Wiko Consult Peering Qos Price and Quality
1. Peering, QoS, and
Price and
Quality Differentiation
J. Scott Marcus, Director
IDATE, Montpellier, 16 November 2011
0
2. Peering, QoS, and Price and Quality Differentiation
• Peering, transit, Internet access
• Quality of Service (QoS)
• The economics of discrimination
• Traffic, costs, prices
• Concluding observations
1
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
3. 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 providers, and 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.
2
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
4. Peering, transit, and Internet access (2)
Peering Peering
ISP A ISP B ISP C
3
ISP A2
ISP B1 ISP C2
ISP A1
ISP C1
ISP A1a ISP C1b
ISP C1a
ISP A1b
ISP A2
3
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
5. 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. 4
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
6. Peering, transit, and Internet access (4)
Shortest exit routing
Backbone ISP
New York
Peering Chicago Peering
Point San Francisco
Washington DC Point
Los Angeles
Atlanta
Dallas
Web Site
Backbone ISP
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.
5
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
7. Quality of Service (QoS) (1)
General Applications Gaming Applications
Stringent Latency Requirements
Stringent Latency Requirements
high
Video
high
VoIP
Telephony
sports, racing, shooter
Channel Surfing
Audio role-playing
IRC Streaming
Video
cards,
Streaming
Browsing, board games strategy
Shopping,
low
Banking
Email
low
FTP
low high low high
Bandwidth Requirements Bandwidth Requirements
• Real time bidirectional audio or video: stringent requirements
• Email: liberal requirements
• Streamed audio and video: liberal requirements (channel surfing?)
6
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
8. Quality of Service (QoS) (2)
• 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.
Source: Cisco (2011).
7
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
9. Quality of Service (QoS) (3)
M/G/1 Queuing Delay (155 Mbps Link)
350.00
300.00
Coefficient
of Variation
250.00
0.00
Wait Time (microseconds)
0.50
200.00 1.00
1.10
1.20
150.00 1.50
2.00
100.00
50.00
-
0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90
Utilization (rho)
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)
8
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
10. Quality of Service (QoS) (4)
• Per-hop delay, even at 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
9
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
11. Quality of Service (QoS) (5)
• 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.
10
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
12. Quality of Service (QoS) (6)
• 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.
11
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
13. Quality of Service (QoS) (7)
• 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.
12
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
14. Quality of Service (QoS) (8)
• 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.
13
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
16. 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.
15
17. Economic foreclosure
• When a producer with market
power in one market segment
attempts to project that market Google
power into upstream or
downstream segments that
User
would otherwise be competitive, Yahoo
Commercial Broadband
ISP ISP
that constitutes economic
foreclosure. Bing
• Foreclosure harms consumers,
and imposes an overall socio-
economic deadweight loss on
society.
16
18. Two-sided markets
• The Internet can be thought of
as a two-sided market, with
network operators serving as a Google
platform connecting providers of
content (e.g. web sites) with
User
consumers. Yahoo
Commercial Broadband
ISP ISP
• Under this view, some disputes Bing
are simply about how costs and
profits should be divided
between the network operators
and the two (or more) sides of
the market.
17
19. 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?
18
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
20. 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.
19
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
21. 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.
Monthly cost of a bundled DSL
broadband/voice service (BNetzA 2009)
Wholesale price ULL Provision
6% Monthly rental ULL
21% Cost of DSLAM
Cost of splitter
31%
Transport in concentration network
9%
Transport IP backbone network
Collocation
3% 7%
Usage dependent cost telephony
6% 14%
Customer acquisition, maintenance, common
cost, billing, bad debt
3%
Source: German BNetzA (2009).
20
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
22. Traffic, costs, prices, and profitability (4)
Fixed Data Traffic
40000 Traffic is indeed increasing in both
35000
the fixed and the mobile networks.
30000
Data traffic/month (PB)
Middle East and Africa
25000 Latin America
Japan
20000
Asia Pacific
15000 North America
Central Eastern Europe
10000
Western Europe
5000
0
2006 2007 2008 2009 2010 2011* 2012*
Mobile Data Traffic
1400
1200
Source: Cisco (2011), WIK calculations.
Data traffic/month (PB)
1000 Middle East and Africa
Latin America
800
Japan
Asia Pacific
600
North America
400 Central Eastern Europe
Western Europe
200
0
2006 2007 2008 2009 2010 2011* 2012*
21
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
23. Traffic, costs, prices, and profitability (5)
• However, the rate of growth in percentage terms is declining over
time.
Source: Cisco (2011), WIK calculations.
22
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
24. Traffic, costs, prices, and profitability (6)
20,000 $25.00
18,000
16,000 $20.00
Here we have the shipment
14,000
12,000 $15.00
quantities in Mbps and the price
10,000
8,000 $10.00 per Mbps (USD) for high end
6,000
4,000 $5.00 routers and for long haul DWDM
2,000
0
2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
$0.00
optoelectronic equipment.
Worldwide high end router capacity shipped (Mbps)
Price per Mbps (USD)
These are among the key cost
drivers for Internet core and
7,000 $7.00 aggregation networks.
6,000 $6.00
5,000 $5.00
The growth in shipments generally
4,000 $4.00 tracks the Cisco projections.
3,000 $3.00
2,000 $2.00
The growth in shipment volume
1,000 $1.00 does not equate to a growth in
0
2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
$0.00
costs, because the decline in unit
Worldwide long haul DWDM capacity shipped (Mbps)
Price per Mbps (USD)
costs is nearly in balance with it.
Source: Dell‟Oro (2011), WIK calculations.
23
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
25. 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.
2,000
1,800
1,600
1,400
1,200
1,000
800
600
400
200
-
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
Source: Dell„Oro (2011), Cisco (2011), WIK calculations. 24
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011
26. 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.
25
Peering, QoS, and Price and Quality Differentiation: IDATE, Montpellier, 16 November 2011