Transcript Marconi - Columbia University

Internet Telephony –
completing the transition
to IP-based
communications
Henning Schulzrinne
Dept. of Computer Science and Dept. of Electrical Engineering
Columbia University
"World Wide Web Redux" -- November 8, 2002
Overview






Communications modes
What makes IP telephony different?
How long will it take to displace POTS?
Events as new service enabler
Making services programmable
IP telephony for emergency
communications
Communications services
traditional
Internet
synchronous, one-way
TV, radio
streaming media
synchronous, interactive
telephone
Internet telephony
"plesiosynchronous"
trunked radio?
instant messaging +
presence
asynchronous, senderdirected
fax, mail
email
asynchronous,
reeceiver-directed
library, CD ROM, phone
announcement
web
What is Internet telephony?
PSTN phones
soft phones
Ethernet phones
VoIP protocol architecture
[email protected]
[email protected]
IP
SIP
proxy
SIP
TLS
TCP
G.711
RTP
UDP
SIP
TLS
TCP
IP
IP
IP
audio
signaling
SIP
TLS
TCP
G.711
RTP
UDP
SIP
TLS
TCP
IP
IP
IP
VoIP as natural evolution

through 1980s: signaling and voice in same
circuit



special signaling tones 
toll fraud, very limited services, slow
1987--: special-purpose packet switched
signaling network (SS7)  out-of-band
signaling


separate physical circuits (64 kb/s to 1.5 Mb/s)
1996--: packet signaling + packet media


physically in-band  higher speed
logically out-of-band
VoIP motivations

Bypass LEC charge


cheaper international calls




VoIP most often invisible as prepay calling cards
effect: Panama outlaws IP telephony
similar to call back in the 1990s
cheaper trunks between PBXs


17c/min in 1984, 0.5c in 2002 (17c in China 2002…)
aggregation into single PSTN termination
new services



multimedia conferencing
integration with Internet services (web, email, presence)
user programmability
Two philosophies for voiceover-IP

carry existing voice services without anybody
noticing




Ethernet is cheaper switching fabric
can share same data pipe
at best, subset of PSTN services
make telephone services just another
Internet service




integrated with email (forward call to email)
web ("click-to-dial")
end system intelligence
same identifier ([email protected] and
sip:[email protected])
IETF VoIP architecture choices

Mobility is not just for wireless:





terminal mobility: change network location
personal mobility: change devices, keep name
session mobility: move sessions to new device
service mobility: services migrate across devices
SIP identifiers are flexible:



one identifier, many devices
one person (or function), one or more identifiers
identifiers are plentiful, cheap and permanent



independent of provider, device or geographic location
authentication, not identifiers used for privacy
Proxies are service-transparent  new services can
be introduced at the edges

e.g., IM & presence added without proxy changes
Technology evolution of PSTN
100
90
80
70
60
electromech
analog
digital
50
40
30
20
10
0
1980 1985 1987 1990 1995 2000 2001
SS7: 1987-1997
VoIP statistics
Some perspective…

Data volume >> voice volume




Netflix DVD rental: 1,500TB/day


AT&T: data passed voice in 1998
now 5x the volume
not true for local calls (LANs?)
Internet: 2,000TB/day
Total US revenue (in $B):
local service
wireless
long distance
Internet (consumer+business)
121
65
109
13
Two views of the future:

IP everywhere




(Vo)IP on cell phones  3G/4G
Internet-based radio and TV
core transport: IP over optical
IP at the edge


(Wireless) Ethernet in LANs and home
edge routers meshed with optical
wavelengths
Programming services


Web success  dynamically generated content, not (just) static
pages
Content creation: small set of specialists  people with other
things to do


similar to audio and video recording
Service creation for IP telephony


from few thousand Lucent & Nortel programmers to every sysadmin
often uses XML as framework for programming languages


dubious, but designed to be written by machines from higher-level
specifications
examples:




sip-cgi for using scripting languages (Perl, Tcl, Python, …)
Java SIP servlets
VoiceXML for voice services
CPL and LESS for call routing and handling
Event notification

Missing service glue:




network management
alarms – "water in level 2"
email alert
geographic proximity alert



media interaction  DVR



"start of show postponed by 30 minutes"
"semantic SMS"
have (ab)used email and polling



"friend Alice is in the area"
see geopriv work in the IETF  location object with embedded security
and privacy policy
email also incurs polling delay
events are typically infrequent  overhead (wireless)
can build services one-by-one  generic platform for quick
service creation
Event notification
alarms
IR detector
temperature
1000
100
10
video
process
control
audio
1
email
0.1
event
interval
polling
SIP events
RTP
0.01
Controlling devices
VoIP for emergency
communications

Easier to re-route calls to gateways far away
from disaster area
NYC



IP
Yale U.
can quickly set up wireless point-to-point links
gradual degradation: 64 kb/s voice  5.3
kb/s voice  signaling only ("Subject:
We're ok")
work in progress to grant priority to
government emergency communications
Challenges

QoS as classical topic since 1990's

but almost no deployment



really, just the short-term version of reliability
Reliability

Internet: 471 min/year (Labovitz et al.)






does not count numerous long-loss episodes
PSTN: 5 min/year (Kuhn)
BGP routing recovery time often several minutes, up to 15'


technical and business complexity
SONET = 50 ms fail-over
Emergency services ("911")
Complexity due to interworking, address shortage
Walled gardens – 3G  no service competition
Security (privacy, DDOS attacks, spam)
Conclusion

Not just an efficient & cheap means of
transporting the same old voice bits



but different spreading mechanism than
web (displacement!)
from vendor/carrier services to usercreated services
transition time of 1-2 decades