www.netnod.se
THE HIDDEN(?) ARGUMENTS FOR FIBRE
Patrik Fältström
Head of Research and Development
Netnod
1
www.netnod.se
35 years ago...
We had one telco
They had some services
Provided TDM based communication
They sold the end equipment
Most fascinating service was call forwarding when there was no answer
The telco was responsible for everything, and legislation was written to target only them
And, they where owned by the government
2
www.netnod.se
20 years ago
We started to get competition
We got packet based communication
First political decisions where made that said that competition was to be enabled
•Number portability
•More than one company selling phones
•More than one cellphone provider
3
www.netnod.se
4
Network in Sweden December 1989
Cisco and µ-vax together with Vitalink bridges created long distance connections
Star-shaped network (64kbps links), with multi-port transceivers as local “LAN” segments
Connection via 64kbps satellite to JvNC in US and to Amsterdam
www.netnod.se
All connections to NSFNet
“Default Network” was pointing at NSFNet
5 connections over the Atlantic: Stockholm, Amsterdam, So!-Antipolis and Pisa
4 large networks: NorduNet, EUNet, Switch and Garr
5
Networks in Europe
December 1989
www.netnod.se
Today a di!erent world
Many telcos
Competition regarding new services
Not only “telephony” uses telco equipment
Internet has taken o"
With Internet, global reach at zero cost
Globalization is here
6
www.netnod.se
Computers and Internet
Everything is in the future a computer, a networked computer of course!
At its simplest your TV, your phone, your address book, your agenda, your micro-wave, you car, your... and your laptop are all networked computers
The Internet belongs to all of us - or at least we all own a bit of it
Each of us has our own personal Internet and some of it we may choose to share
Increasingly each of us runs part of the infrastructure
7
www.netnod.se
Convergence?
Information = Software
Anyone can create Information
Anyone can create Software
Anyone can distribute Information
Anyone can deploy Services
8
www.netnod.se
Convergence?
Historically we know who can create information
Historically we know who can create software
Historically we know who can deploy services
Now anyone can deploy services
9
www.netnod.se
My piece of the Internet?
When a person or organization connect to “the Internet”, the network and services provided end up being a piece of the Internet
Protection (and robustness) start at home
You have a lock on your door, and do not ask road authorities to keep burglars out!
10
www.netnod.se
Historically we had this view
ServiceTransmissionWire
ServiceTransmissionWire
ServiceTransmissionWire
11
www.netnod.se
Today, Internet architecture
Internet
Service
Wire
12
www.netnod.se
Old and new world
Telephony, Cable TV, Satellite, Mobile
•Buy connection from one provider
•Then buy additional services from provider
Internet
•Buy connection from one provider
•Then buy additional services from anyone
13
Old
New
www.netnod.se
Old and new world
Telephony, Cable TV, Satellite, Mobile
•Buy connection from one provider
•Then buy additional services from provider
Internet
•Buy connection from one provider
•Then buy additional services from anyone
14
Old
New
www.netnod.se
We have introduced abstractions!
Internet
Service
Wire
Independence!
15
www.netnod.se
Internet eco system...
ApplicationDevelopers
Broadcasters
Advertisers
Content Aggregators
Retailers
Cloud Services
Network ClientDevices
16
www.netnod.se
What drives productivity?
Technological change and other factors
ImprovedLabor Quality
Capital Investment
52%
37%
11%
17
www.netnod.se
Winners and loosers
Increm
ental V
alue
Add
Winner
Loser
Electricity
Telephone
Internet
Mobile telephone
Inven3on
highvalue add
mediumvalue add
lowvalue add
INNOVATION
Poten3alforcompe33veadvantage
Act later
NG ICT Infrastructure
Television
Time
18
www.netnod.se
Internet gives the abstraction layer we want...
Given one have Internet Access, one can get services from anyone
In reality because of this, we have three important layers:
•The passive infrastructure
•The Internet Access
•The services - which are the ones end users want
What do each one of these cost, where should we spend money?
19
λ λλ
www.netnod.se
Packet based networks?
In each network element you see an M/M/1 queuing mechanism
Relatively simple to show how more e"ective packet based systems (M/M/1/K with combined poisson and exponential distribution) are than traditional TDM
20
0 1 2 n-1 n n+1
λ
μ μ μ μ.....
λ μ123.....
www.netnod.se
Degraded functionality in a packet based network
Functionality
StressPoint where network collapses
21
www.netnod.se
Two di!erent questions that should not be mixed:
22
How to communicate when communication network has collapsed?
How to move the point z as much to the right as possible?
www.netnod.se
We have a few di!erent problems:
Do we know what will break?
Can we increase ability for the system to withstand stress?
What do we do when things breaks?
Can we minimize the amount of unwanted tra$c?
23
What is unwanted tra!c?
www.netnod.se
Time
Whatever
Internet ends here
We are here now
6 months to 5 years
How often have you not seen pictures like these?
www.netnod.se
1921 - Sweden
1998 - Canada
2005 - Sweden25
www.netnod.se
Example of a problem: IPv6
26
www.netnod.se
Example of a problem: IPv6
Content
User
ISP
Device
“A deadlock, stalemate, impasse; a roughly equal (and frequently unsatisfactory) outcome to a con%ict in which there is no clear winner or loser,”
Where is the content?
Where is the network?
Do I pay less or get new applications?
NAT’s are good. RFC1918 gives me security, and IPv4
address runout is my ISP’s problem.
The network is not ready, users don’t care and I don’t want to risk a poor end-user
experience today for potential gains tomorrow
Enterprise
27
www.netnod.se
We move into third phase of Internet
1980–1995
1995–2010
2010–2025
Era of deregulation and competition, Internet arrives
Early days of Internet, service providers, social networking, mobile Internet
Internet takes o"...
28
www.netnod.se
0
1500
3000
4500
6000
1981 1985 1990 1995 2000 2005 2010 2015 2020 2025
5448,5
2724,3
1362,1
681,1353,3
93,04,90,30,00,0
The dot com boom!
Devices(millions) 29
© 2011 Cisco and/or its affiliates. All rights reserved. Cisco Public
Video Will Become the Predominant Traffic
Gopher, FTPWWW
P2P
1993-19951995-2000
2000-20132013-2025
2025+
1995: Web Overtakes Gopher, FTP2000: Peer-to-Peer Overtakes Web2013: Video Content Overtakes Peer-to-Peer2025: Video Communication Overtakes Video Content
VideoContent
VideoCommunication
www.netnod.se
31
www.netnod.se
32https://labs.ripe.net/Members/fergalc/ixp-tra!c-during-stratos-skydive
www.netnod.se
One customer...mid october...
33
www.netnod.se
34
www.netnod.se
35
www.netnod.se
36
"Normal" changes can create problems!But only if one is not prepared...
QoS Does Not Help!
www.netnod.se
But what helps?
37
www.netnod.se
But what helps?
38
Existence of dark #bre as wholesale product!
www.netnod.se
But what helps?
If we look at the investments to be made, passive infrastructure investments can (and should) have long pay pack time, speci!cally ducts and !bre -- that can be used by everyone
In each geographical area, having more than one provider of passive infrastructure is not e"ective (who want to be #2?)
For example, each area (city) can be divided in four areas:
A.Where investment in "bre give good economical gain
B.Where investment in "bre pays back, but not more
C.Where investment in "bre must be subsidized (from A?)
D.Where investment in "bre is not possible without economical loss
39
www.netnod.se
New ideas and investments?
But...we have historically many di"erent communication mechanisms:
•Wikipedia talks about more than 80 di!erent protocols
•Many of them are historic, geographically local serial based ones
•Today people people use IP, ethernet, "ber or radio
Today, passive infrastructure is shared
•Radio towers and masts
•Fibre
•RJ45 / Cat5e
40
www.netnod.se
Example, a new time distribution mechanism
Time and Frequency Transfer in an Asynchronous TCP/IP over SDH-network Utilizing Passive Listening
http://tycho.usno.navy.mil/ptti/ptti2005/paper98.pdf
Figure 1. Schematical setup at a router in an TCP/IP network.
III. TECHNICAL ARRANGEMENT A schematic view of the proposed setup at one router in a
TCP/IP network is shown in figure 1. We couple 1% of the outgoing optical power, and 10% of the incoming power, to extract the output and input timing signals from the traffic. This is a negligible part of the transmitted power, since an operational system typically is designed with more than 50% power margin. The deflected optical signal is then converted to an electrical signal using an opto/electric converter (o/e). The converter must be 10 times more sensitive than the o/e used within the router, but nevertheless this is a commercially available device. This signal is fed into a Header Recognizer which is based on standard SONET/SDH circuit components in conjunction with a Field-Programmable Gate Array (FPGA), and has been developed together with InformAsic AB. The Header Recognizer generates a pulse of length 25 ns every time it receives a sequence of 384 bytes of the SDH frame alignment bytes A1 and A2, see Figure 2. This pulse triggers the start of a time interval measurement in the time interval counter (TIC). A similar pulse generated from the local clock indicates the end of the time interval. Such time interval measurements will be generated at a rate of 8 kHz, i.e., the rate of the incoming
frames. For the clarity of the figure, the measurement of the outgoing traffic is not included. However, an equivalent setup will be installed also there, in addition to the in- and outputs of the other fiber lines connected to the router.
IV. TIME TRANSFER PRINCIPLE Figure 3 and Figure 4 show the main principle behind the
time transfer between two routers A and B equipped with clocks providing the time CA and CB respectively. At the outgoing fiber, the time of a start of frame, TAo, is detected in the SDH signal. Analogously the time of a start of frame, TBi, is detected in the SDH signal at the incoming signal at router B. The same principle is then used for the signal in the opposite direction. There will thereby be four time intervals, associated with each point-to-point transmission. By comparing the time, CA, by the clock at router A with TAo (as depicted in Figure 3), we can measure the time difference 'CAo. Hence defining 'CAo, 'CAi, 'CBo and 'CBi as:
� AAoAo CTC � ' � ����
� AAiAi CTC � ' � ����
� BBoBo CTC � ' � ����
� BBiBi CTC � ' � ����
Furthermore, the travel time in the fiber, approximately 5 µs/km, is taken into account through F(t) such that:
� � �tFTT AoBi � � ����
Figure 2. The header recognizer generates a pulse with the length of 25 ns every time it receives a correct sequence of A1 and A2 bytes.
909
41
www.netnod.se
Summary...
•Passive infrastructure is no longer tied to services
•Passive infrastructure is to be shared•Owner of passive infrastructure decide themselves if they get competitors
•In Sweden, if ~1500 households can be reached from one node [with active equipment] providers of transmission compete on being the "rst•Payment is "xed fee per household per month, regardless of distance
•Competition is the future, and we get competition via abstraction•Historically, the incumbent came with a few * and # feature per year
•Today, in Apple AppStore alone, we got 339k apps during 2012
•Given one have Internet Access, one can get everything•Triple play is dead as a technology, but will always exist as contract/bundling
•Investment in "bre is a good investment!•STOKAB 2011: Net sales €73M, Investments €65M, Pro"t before tax €19M
42