Transcript
Evolving the Internet: Challenges and Opportunities Mark Handley Professor of Networked Systems University College London
The Power of Legacy
(Pictures courtesy NASA, Darlington Railway Centre)
Brunel’s 7-foot gauge. More stable, faster, more spacious. But, better technology often fails to win. Network effect: Unloading cargo to transfer between railways was too expensive.
Network Effect Metcalfe’s Law: The utility of a telecommunications network grows with the square of the number of users.
Picture by Derrick Coetzee
The Challenge of Change
Why do some Internet technologies succeed and some fail? How does technological change come about? Revolution vs Evolution What happens when change fails?
Picture courtesy Open University/BBC
Key Challenge
Is it possible to change the Internet architecture in a planned way, so as to achieve long-term goals?
Internet map, 1999
Source: Bill Cheswick, Lumeta
Internet Time
http://www.stupid.com/stat/RTCL.html
Things change so fast… YouTube MySpace BitTorrent Skype Joost
What’s the new new thing?
Digital Convergence: One Network Connecting Everyone
Telephone Network Data Network TV Network Mobile Telephone
Global Internet
How do other networks change?
Evolving Networks (1) Rail Network
Two basic services (move people, move trucks). Switch from steam to electric doesn’t change the service. Interconnect steam and electric at stations by simply changing the engine.
Evolving Networks (2) Traditional Telephone Network
One basic service. Switch from operators to direct dial. • Old phones could still call the operator. Switch from analogue to digital. • Can interconnect analog and digital at gateways because the service is well understood.
The Internet is Different
The Internet is Stupid
It doesn’t know what problem it solves. 80% of the functionality for 20% of the cost.
The net doesn’t have any embedded knowledge of services. It can support new unknown services. It can’t tell when it is working.
Different
In 1992 we didn’t see the web coming. By 1995 it was 50% of the traffic. In 1999 we didn’t see Napster coming. By 2002 peer-to-peer file sharing was 50% of the traffic. We won’t see the next killer app coming either. Need to design the network to be flexible.
Revolution or Evolution?
Revolution is the norm for Internet applications. Web, peer-to-peer, Internet telephony. The new new thing…
Revolution: new link technologies appear every few years. WiFi, WiMax, Metro-ethernet, Optical switching, Passive Optical Networking (PON). Often re-use existing copper or fibre in radically new ways.
Evolution is the norm for the core of the Internet.
Change
Huge innovation in applications
email WWW phone... SMTP HTTP RTP... TCP UDP…
Ossification of the core protocols
IP ethernet PPP…
Relentless evolution of the underlying technology
CSMA async sonet... copper fiber radio...
Evolving Road Networks
Road Networks
QuickTime™ and a TIFF (Uncompressed) decompressor are needed to see this picture.
Changing the Core of the Net.
QuickTime™ and a TIFF (Uncompressed) decompressor are needed to see this picture.
1st Jan 1983. Flag day. ARPAnet switched from NCP to TCP/IP. About 400 machines needed to switch. We won’t get to do this again. Future changes must be incrementally deployable, and provide financial benefits for early adopters.
Growing pains… 600
Hosts (Millions)
500
Number of computers on the Internet
400 300 200 100
Source:Internet Software Consortium
n08 Ja
n06 Ja
n04 Ja
n02 Ja
n00 Ja
n98 Ja
n96 Ja
n94 Ja
n92 Ja
n90 Ja
n88 Ja
n86 Ja
Ja
n84
0
Development Cycle We need this new feature to keep our network functioning
Quic kT i me™ and a T IFF (Unc ompres s ed) dec ompres s or are needed t o s ee thi s pi c ture.
Quick Time™a nd a TIFF ( Uncomp res sed) deco mpre ssor are n eede d to s ee this picture .
Qu ic kTi me™ a nd a TIFF (U nc omp res se d) de co mpre ss or are n ee de d to se e thi s p i cture .
Here’s a solution. Let us know how it works.
You are here
You are here
You are here
You are here
The story so far
The overwhelming power of legacy constraints. Network effect weights change towards incremental change in the core.
No-one is in charge.
The Internet is more general than other networks: Huge range of possible applications. No well defined service to gateway between old and new.
Very fast growth leads to short-term planning. Hill climbing without a map.
A Tale of Two Technologies
A Tale of Two Technologies
IP Multicast Routing and Addressing. How to turn the Internet into a broadcast medium. Still not deployed worldwide, despite the rise of Internet television.
Signalling for multimedia flows: SIP (Session Initiation Protocol). • Global standard for how to make a call in the Internet. • Now a multi-billion dollar industry. • Replacing the old telephone network.
Why does SIP succeed and Multicast not, when there’s a clear need for both?
The Story of IP Multicast IP Multicast (the theory): Great for application writers and content producers: send one packet, get it delivered to millions of receivers. Great for ISPs: pay upstream provider for one packet, satisfy many customers.
Sender
The Sad Story of IP Multicast
We botched it! Original IP Multicast model is very general. Senders just send to a group address. Receivers ask to receive traffic sent to a group address. Network does magic.
Magic is hard to debug. IGMP + PIM-SM + BGMP + Malloc = high deployment costs.
Eventually we realized it wasn’t going to happen.
The Sad Story of IP Multicast
New service model: Senders send to group address. Receivers tell network which senders they want to receive.
Much much simpler. Still a revolutionary change though. IP Multicast (the reality): No ubiquitous deployed service. Applications couldn’t depend on it, so had already implemented alternatives. No obvious customers, so not worth investing in the network infrastructure when the investment won’t clearly pay off.
The Curious Story of IP Multicast IP Multicast is becoming a success!
Commonplace in corporate networks. Same company runs the applications as pays for the network.
Becoming commonplace in consumer ISPs. ISPs are becoming Cable TV companies: • Triple-play (Phone, TV, and data, all over IP to the home). • Money from services; data alone is unprofitable.
But little deployment of multicast between ISPs because this would undermine their own TV services.
The Story of SIP
Video conference, 1994
The Story of SIP (1997) Two academics: Needed a way to invite people to join video conferences. No existing standards designed for global Internet use. Rolled our own simple protocol in the style of HTTP, but including proxies to support user mobility. Industry giants: Had existing video telephone standards. Wanted to extend this to corporate LANs. Result: H.323, picked by Microsoft and Intel to be the solution for Internet video conferences.
What chance two academics against Microsoft and Intel and the power of legacy?
A Little Stubbornness
IETF Multiparty Multimedia Session Control working group. Provided an audience with a preference for Internet style protocols. • SIP: easy to understand. • H.323: must read 500 pages of only partly-relevant specs.
We simply ignored H.323 and Microsoft and Intel, and noone stopped us.
SIP: User Location
There are two basic ways to do user location:
Have a distributed directory. Lookup during call routing Pictures from Cisco, 3
SIP: User Location Option 1: Distributed Directory for User Location. Lookup user's location in directory. Address a call to that location.
Problems: Agreement on directories (X.500 not a resounding success) Privacy issues: can track user location without calling Directory them. Where is Mark? Mark is at work
Call Mark at work
SIP User Location Option 2: User Location while Routing the Call.
SIP: User Location while Routing Advantages: Privacy. Different places can use different mechanisms for user location. Decentralized local administration.
SIP can be deployed in many different ways, and they can all talk to each other.
Data is the New King Data rate
From here on, carrying phone over IP looks inevitable
Circuit-switched phone traffic Internet data traffic
Time
Proxies make the difference 1998: MCI are starting to investigate Internet telephony. It is obvious telephony will go this way eventually. How can phone companies make money?
Henry Sinnreich likes SIP because he sees that MCI can run the SIP proxies in their network. By inserting themselves into the signalling path, this provides a way to bill for service. Not what we intended proxies to be used for!
But, suddenly lots of telcos are interested in SIP.
Design for Tussle When technologies succeed, it is because people can see a way to make money. Some technologies solve a problem really well, but are brittle. Some technologies are flexible - they may be less optimal, but don’t break when you bend them to new tasks. Technology designers rarely understand how their technology will get used in practice. Different players pull the technology in opposite U.S. Navy photo by Photographer’s Mate 2nd Class Ryan Child
Timing Matters
When stresses build, eventually something happens. Revolutionary change is not always well-planned.
The technology that is in the right place at the right time wins. Too early, there’s no perceived benefit. Too late, another technology already benefits from the network effect.
San Francisco, 1906
What’s changing today?
The Internet is Stupid
It doesn’t know what problem it solves. 80% of the functionality for 20% of the cost.
The net doesn’t have any embedded knowledge of services. Telephone This is a good thing. Network Data It can support new services. Network TV Network Mobile Telephone
Global Internet
New services are not always good… Date: Sat, 10 Nov 2007 09:57:43 -0500 (EST) Subject: INVESTMENT PROPOSAL From: SENATOR PATRICK OSAKWE DEAR FRIEND, I AM HON. PATRICK OSAKWE, FROM NATIONAL ASSEMBLY OF THE FEDERAL REPUBLIC OF NIGERIA. CAN I ENTRUST HUGE MONEY ON YOUR CARE FOR FUTURE INVESTMENT? I NEED YOUR FULL NAME AND TELEPHONE NUMBERS(CELL) TO SPEAK WITH YOU. REGARDS, SENATOR PATRICK OSAKWE
Worms, Viruses, and Denial-of-Service Attacks
QuickTime™ and a TIFF (LZW) decompressor are needed to see this picture.
Normal daily traffic
5:30am GMT 24th Jan 2003 Slammer Worm
The Challenge of Security
How can we mitigate security threats without sacrificing the future? How to provide robust network services in the face of attack? How to preserve the ability to innovate?
Extrapolation of current trends does not bode well....
The Stresses are Building
Increased reliance on the net Phone, TV, etc
Core of the network is largely unchanged since 1993 Security problems Routing scalability Congestion management Address exhaustion
The limits of evolution?
http://www.gocomics.com/luckycow/
You are here
Is it possible to change the Internet architecture in a planned way, so as to achieve long-term goals? (or is it only possible to patch the pieces repeatedly until it gets too expensive and unreliable, and eventually something better comes along and replaces it?)
The Role of Networking Research
Which mountain should we climb? Innovations needed for the ascent. To be relevant, good ideas not sufficient: Need good timing. Need good communication with industry. • What problems will they have? • Demonstrating relevance of your ideas is hard but essential. Design to allow your technology to be deployed in ways you don’t expect.
On demonstrating relevance…
Above all, build real systems. The HEN network testbed at UCL
