future internet developments: evolution or flag days john c. klensin, ph.d. apan august 2005 © john...
TRANSCRIPT
Future Internet Developments: Evolution or Flag Days
John C. Klensin, Ph.D.APAN August 2005
© John C Klensin, 2005, All rights reserved
2005.08.23 2
Transitions in Network Technologies
• Parallel development andcompatible operation
• Simultaneous– Shutdown of one service – Starting replacement
• Latter is “flag day”– Dangerous and nasty– When needed?– How likely for the Internet?
2005.08.23 3
Reviewing Network History
• Disruptive and non-disruptive changes
• Smooth evolution, evolutionary jumps, discontinuous shifts
• Transition and “flag days”
2005.08.23 4
The Norm in Distance Communications
• The norm has been either– Evolution within a network model
• Either within a technology• Or with one technology running in parallel to
another until it fades away
– Shutdown of one, start of another• The “flag day” model• Usually occurs due to political pressures/
decisions, not technology
2005.08.23 5
Exploring a Hypothetical Case
• Suppose it were decided to shut the PSTN down, worldwide, on a single day, leaving only VoIP
• Would not be a true flag day – VoIP runs today, so little risk
• Interesting questions…
2005.08.23 6
Hypothetical Case: Questions
• Some questions– Why try to think about it globally?– Who would decide?– Why would they decide?– Who would pay attention?
2005.08.23 7
Hypothetical Case: Environment
• All of the choices are about politics or economics, not technology– If a shutdown decision actually ever occurred,
• Carrier by carrier• Country by country
– Anyone recently sent • A telegram • A package by horse-borne messenger?• A smoke signal?
they did not disappear by global policy
2005.08.23 8
Four Examples of Superceding Technologies
• Couriers on horseback to trains
• ARPANET to Internet Protocols
• Host tables to the DNS
• Navigation and exploration tools– Gopher, Archie, WAIS, etc. and– The Web
2005.08.23 9
Couriers on horseback to trains
• Relay couriers essentially disappeared overnight on train routes
• No planning or policy, just obsolete
• No flag day either: no technical reason the horses could not keep
running
• Incidentally, no “convergence”
2005.08.23 10
ARPANET to Internet Protocols
• The January 1, 1983 “Internet Flag Day”• All NCP services and transition between NCP
and TCP/IP shut down• Not really a discontinuous flag day:
– Services ran in parallel for a while– TCP/IP was well tested
• Getting rid of NCP– An economic and policy decision
• And forcing TCP/IP adoption– A policy and economic decision
2005.08.23 11
Host Tables to the DNS
• DNS installed, host table continued to be distributed
• Long series of deadlines about host table distribution shutdown
• Even after host tables were not distributed, not all hosts ran DNS
• Full conversion caused by rising inaccessibility, not policy decisions– Pain of not converting exceeded pain to convert.
2005.08.23 12
Navigation and exploration tools
• Several ideas and protocols– Early ideas and techniques– Gopher, Archie, WAIS, etc. and
the Web• Evolution
– Coexistence until no one cared any more– Then disappearance
• No global policy decision, local economic decisions
• Incidentally, some good capabilities got lost in the process
2005.08.23 13
Few Cases Require Real Flag Days
• A Good Thing because– Real flag days are a nasty business
• Service outages– Short if successful– Catastrophic if not
• Coordination problems• Questions of authority in non-centralized network
• Shutdowns of parallel/ redundant services• Can be fairly painless• Still require careful planning
2005.08.23 14
Forcing Transitions
• Many good operational reasons to force transitions
• Parallel operations rarely a problem– Applications in parallel– Or infrastructure in parallel
• Same application, parallel infrastructure– Rarely works as well as we hope– Strong incentives to get the transition finished
2005.08.23 15
The Forcing Functions
• Economics, Scale, or Policy
• Interesting examples…– New email requirements for spam-fighting?– IPv6 ?
• But– neither requires a flag day– either would work better if everyone
transitioned
2005.08.23 16
The Old Questions Come Back
• The policy ones…– Who would decide?– Why would they decide?– Who would pay attention?
• A “no one gets to run the old protocol after…” decision– Would effectively create two networks for the service– Business opportunity for gateway/ conversion
providers
But such conversions often do not work well
2005.08.23 17
Better to Attract than to Mandate
• Requires some patience
• No overnight cures
• But no overnight disasters, either.
2005.08.23 18
IPv6 Example: A New Base
• Appeal should be to – New users and networks– New applications– New areas– For them
• Not already happy with current environment – not using it• No conversion costs: direct to IPv6
• Not trying to force IPv4 conversions until– There is lots of IPv6 out there– Ideally supporting new services
2005.08.23 22
When Do We Need a Flag Day?
• Incompatible protocols that cannot co-exist on the same network
• Most common case:No way for server to tell which protocol is in use
• Usually the result of poor or indifferent design• Getting from IPv4 to IPv6
– Common first part of packet header with version number
– Could have gotten that wrong, but it would have been dumb
2005.08.23 23
Possible Example
• Major change to routing structure– Not just new version of BGP or equivalent– Difference in how we think about routing packets
• But, even then…– Routing within a network is different from routing
between them– Many features of the current model would make it
easy to isolate changes• Roll out new model, one network at a time• Lots of little-network flag days, no global Internet one• And that is the worst case
2005.08.23 24
Getting this Right
• We are good at evolution without flag days Consider Internet Mail– 1982 Limitations
• No multimedia or structured content• ASCII-only• No delivery notifications or security/privacy facilities
– 2005: Those limitations, and others, removed– 2006: Internationalization of addresses and
headers ??
• All without incompatible infrastructure changes, much less flag days
2005.08.23 25
Who Might Like a Flag Day
• Beneficiaries of incompatibility or chaos• Economics:
– “Everyone should be forced onto my network design, so I can make more money”
• Desire for different policy or management model– “We understand circuit networks better than packet
ones, so the Internet should be a circult network”– “The Internet is hard to control, so the solution is to
break it and replace it with something easier to control”
2005.08.23 26
A Dumb, Edge Network
• Should not be looked to as the solution to– Political problems
• National relationships• Cross-border issues
– Social and legal problems• Content regulation• Spam and other offenses
• Strength of the Internet is in the “dumb network, smart edges” model
2005.08.23 27
Preserving Interoperability is Key
• “Internet Telephony” is getting interesting– The superior model in every way is
• Based on standard protocols and open standards• Multivendor, multiprovider environment with
opportunities for– New entrants introducing new services– Competition to provide good service and innovation
• Standard, PSTN-compatible, numbering model
– But Skype has one advantage…• It works• and it is arguably winning
2005.08.23 28
Conclusion
Flag days are one problem we do not need to worry about….. Unless– Network-killing policy decisions are made and
anyone pays attention– We get in too much of a hurry and make
shortcuts with critical infrastructure– We decide we don not care about a global,
interoperable, network
2005.08.23 29
Thanks to Rob Austein, Vint Cerf, and Dan Lynch for generously and quickly filling in some details of the 1 January 1983 transition that I had forgotten or not known.
2005.08.23 31
Flag Days and Reality
• No foreseeable cases in which global Internet flag days are neededMuch as some might secretly wish for them
• That is good because– A flag day transition would
• almost certainly fragment the network• as some ignored the mandate to switch or• were unable to implement it
– There is no plausible mechanism for requiring and managing one.