General
impressions:
The
atmosphere and energy here were higher than for the Atlanta show, though
probably still less than a year ago.
Introduction of VoIP in Enterprise and Internationally continue to be
the hottest markets, though services remain of intense interest to
everyone. One of the keynote speakers
talked about an evolution of the industry from a focus on transport to one on
services, and I agree that this has been evident. The hot services remain IP centrex/IP PBX, and presence based
services remain interesting, though more interesting media services are
starting to appear. There was
definitely more about video and broadband than in the past.
SIP
remains the protocol of choice for everything, though it’s no longer
controversial. More like a foregone conclusion. SIP doesn’t do everything yet, but almost
everyone talking about communications talked about SIP.
The after
effects of September 11th still show up in a lot of presentations
and discussions. It is interesting that
Lucent, Nortel, Verizon and others are justifiably proud of how their high
reliability products and networks performed and the extraordinary efforts they
made in restoration, but the incident clearly also demonstrated the key
advantages of a well connected IP network with lots of excess capacity. The IP networks quickly routed around
damaged routers and cables, and the excess capacity allowed it to take on
substantially more load.
There was
the usual discussion of applications of every sort, and application platforms
(Java, Parlay, etc.) but in the discussions I was in more focus on what the
value propositions are than the details of how it’s done. Almost a “show me the money” feeling.
Several
keynotes and session presentations talked about the internet as the driving
force and the disruptive technology. I
think maybe that hit home a little harder this time. Many speakers talked about not bending the internet too far
towards a circuit model in order to achieve high quality of service for
voice. Others talked about the
economics of various VoIP applications being significantly better using the
internet (not a private, managed IP network) for transport. Still others talked about the contrasts
between the results of adopting a carefully controlled model of building 3rd
party applications to that achieved by a really open model in which anyone with
an idea can turn it into an application.
I think all of these point to the need to look at a real internet (not
IP technology and “open but protected/managed” interfaces) model as a serious
contender.
What’s
probably as interesting as who was here is who wasn’t or was here with a lower
profile than in the past. That included
big vendors (Lucent, Nortel, Alcatel), Newer companies (Dynamicsoft,
NexVerse/iPVerse, and Carriers (Worldcom, Level 3). There seemed to be more mainstream carriers attending, with
people from Verizon, SBC, Sprint, NTT, and others. Microsoft probably had the largest booth on the exhibit floor,
and debates of Microsoft’s approaches (SALT, .Net, etc.) vs others were part of
many discussions.
Here are
my more detailed notes for the conference.
The
exhibit floor was large and well
attended, with a lot of familiar exhibitors and the usual mix of
startups. In talking with exhibitors I
knew I got the impression that the companies selling systems and products were
seeing better traffic than the companies selling components integrated into
VoIP systems. Maybe this reflects more
carriers and business users in attendance, or maybe it’s a reflection that the
natural customers for components are still consolidating and not buying as
much.
I didn’t
see as much dramatically new this year, but that may reflect the fact that I’ve
seen what is happening. Many companies
exhibited applications and media platforms of various sorts.
Microsoft
had a large booth and an interesting one demonstrating XP, but also
demonstrating some of their new technology initiatives (.Net and SALT), and
partner applications. The SALT
demonstration was quite good.
Several
companies displayed fault resilient computing platforms offering rapid fail
over and recovery. Again, more
incremental improvement than revolution.
SONUS --
63% of the market 50% of the minutes go
through their gateways. Lots of
cutbacks in carrier spending, but not a disaster for them. The long term growth is still doubling every
year (minutes). One analyst said
spending in general is back to 1999 levels and basically expected to stay flat
at this level. (That's all carrier spending, not just VoIP, which is growing).
Some analysts are projecting a rebound "bubble" from pent up
demand. (Comment -- flat sounds right
to me. Prices are dropping, nmber of
people and total number of minutes in all forms of communication aren't going
to grow very fast, so why would you expect double digit growth in the equipment
market?)_
He gave an
example of a VoIP greenfield that worked -- Fusion communication -- LD and
International provider for Japan.
Turned it up very rapidly, over a million customers (first year I
think). Few surprises. It's just plain
old telephony that happens to run over IP.
They have
Hundreds of softswitches and 3 Billions of minutes of use. Millions of ports. (Sounds impressive. Sonus
is pitching their real advantage in this as having real carrier class products,
not just NT boxes with Telephony cards in them, and that the availability of
high performance carrier solutions is the key enabler to make this happen.)
Billions
of minutes of VoIP are great, but we have trillions. We have seen loss of growth and profitability on the voice side
but have not yet seen the growth of profitability for data. It is time to recognize the winners and
losers and focus on moving on to new technology with a view for profit.
IP won the
revolution, that war is over -- Frame Relay and ATM are the losers. MPLS isn't embracing what is new, it tries
to make IP like Frame or ATM. Let's
look at what people like about Frame and ATM, and provide that in an IP way,
not by making IP look like Frame or ATM underneath. Sprint embraced the internet way up front.
We are at
the "Cambrian explosion" of services. New communication services will make it possible for things that
people couldn't do. Companies like Broadsoft and Sylantro have plenty of
innovation, they need an infrastructure to build on -- that's IP, not IP made
to look like something else.
Service
Consistency (never being out of service – which he credited to SONET)
Personalized services (everything is personalized) Dynamic bandwidth Allocation
Total Service integration -- everything is integrated (Comment -- can you
say Feature Interactions?) Multiple concurrent services on the same
platform (not specialized to service) is important. It is also going to be important to extend the network to the
premises (not something you have to get to).
Innovation
– look at the Hula Hoop – It wasn’t technological and made use of existing
technologies to give people something they wanted. (He really cited this as the prototype for innovation in
SERVICES, versus Technologies) We can't
predict what they will be (Comment -- how many hula hoop like things were
invented but never became popular? we
keep expecting innovation to have a high success rate. I think in fact it has a very low and
unpredictable success rate.)
The basic
Message -- this isn't a technology problem.
You need innovaters that understand people to come up with the ideas.
Technology has to support them and support the ability to do this, not support
the old ways of doing things.
How do you
think about Innovation? Ask what does
the world I want to live in look like?
Don't cut it off by a near term time frame because it is too limiting. Should be asking what the world should look
like in 500 years. (Comment --
whenever I think about this kind of question, I ask myself, what would people
500 years ago (Elizabethan era I believe) have thought the world would look
like today? Even this isn't realistic,
because the pace of change is exponential (i.e. more change in next 10 years
than previous 10 years). In this case
it's interesting. Shakespeare wrote 500
years ago, and his works are still a pretty good model of our human
interactions today. We have advanced
lightyears in technology, but hardly at all in the way human interaction and
relationships).
Transport
has become a commodity -- driven to the lowest possible price by aggressive
competition. Only way to make money is
value added services -- you need an infrastructure to support them. The thing
that is neat about SIP is that the first word is "session"
(presumably as opposed to call, connection, or application). Service
architecture is session oriented (Comment -- yes of course, Sprint has done
a lot of work on the TINA model and the TINA model is session based. Of course I'm not sure how consistent the
TINA view of session is with the IP view that drives SIP).
Energy
should not be spent on Convergence any more -- That was 4-5 years ago and
driving the move to IP. Unfortunately
that focus is driving towards a multiplicity of protocols and adaptations (i.e.
bend IP to serve all the existing communication models, not go back to the
fundamentals of what services people really want).
Shrink
wrap versus Network Resident applications -- Consumers want to be able to pick
their choices (like buying a video or a software ap). People working together want everything everywhere to work just
like their office. (Comment --
actually I'd differ on this. Users still want their customizations. People put their own customizations on their
PDAs, mail clients, phones, etc. What
you need is interfaces that clearly separate the personalizable from the "needs
to be standard for universal accessibility", not proprietary interfaces
that leave the user the choice of accepting interfaces they don't like, or
struggling to communicate with others who made other client choices).
Sprint
"did this ION thing a few years back, valuable in teaching lessons about
how service architectures work." Also learned a lesson that they can't do it alone. Sprint is now focusing on solutions and
working alliances (Cisco and Nortel I think were the ones he mentioned). Business isn't about minutes, it's about
providing solutions to telecommunications problems.
Average
life expectancy of a US business is 50-100 years. (Comment -- I presume this means big, established businesses,
because it's a lot less if you include every one. I won't point out how old some of the companies in our industry
really are).
Past VONS
were lots of fun because of the presence of lots of entrepeneurial people with
nice ideas. Past year was brutal with
lots of people leaving the industry. We
did the right thing -- IP won, IP is the future (not sure how this relates to
the comment on the VON audience) He
related an interesting story on the Slinky toy -- An engineer on a ship noticed
a broken spring bouncing along the floor, and turned it into SLINKY.
Enterprises
embracing VoIP -- Cisco has (Worldcom has I think), Sprint is about to do it,
Telcordia (SAIC) is about to do it.
This will make lots of minutes of use on circuit networks disappear. All
intra company traffic will go over internal networks.
(Comment
-- he was a last minute substitute because the planned speaker was attending
the state funeral of the Queen Mother.
He clearly had a talk focussed on standards, which was kind of strange
for a VON).
Standards
have made it possible to do the same things in a lot of places
(interoperability and flexibility)
Customers
don't buy things like Convergence, IP, or
Broadband. Customers buy value
(customer retention, new revenue, cost reduction, etc.) Standards are only useful if they deliver
value. Clear value of standards is to get things to market faster because you
can make things that come from different environments and backgrounds come
together.
Have to
address the problem that you can't do a flash cut to anything. Users justify their capital expenditures
based on what new revenue or other value that they bring in.
What's the
killer application? -- The one people find easy to use. It has to be
intuitively obvious to people how to get the advantage. Example -- QWERTY keyboard -- known to be
not optimal, based on mechanical limits of early typewriters, yet you still see
them on PDAs and Blackberry's.
Why? It's the natural user
interface. (Comment -- this kind of
thinking is also the reason for perpetuation of a lot of bad interfaces as
well.)
Collaboration
-- Look at what users are doing now with Instant messaging, chat rooms,
etc. Make collaboration on the phone
just as easy. (Comment -- everyone's
trying to do this. It is interesting
that I built a system in 1979 for Bell Labs that had an simple visual interface
to phone features much like that we now see done with presence based calling
systems. We published a couple of
papers on it, and certainly didn’t invent the idea. I borrowed it from a couple of people in Bell Labs Holmdel who
were trying to build an intuitive user interface for PBXs, and I doubt it was
original there. Among other things
their implementation had menu based interfaces for doing conferencing,
transfer, and multi-appearance management on a plain old phone with no display
using IVR dialogs. I'm not sure this
would ever have been intuitive to everyone, but it's interesting to see how old
some "new" ideas areally are).
Convergence
is about making this easy -- plug and play, speech enabled, push a button. Vendors need to focus on what the end user
experience will be.
There is
tension between innovation, value creation, and standards. (Standards take time, some innovation
doesn’t have immediate value, etc.
I walked
in a bit late on this and only caught a couple of the talks.
He
presented the Personeta Application server.
This is a Java based platform built to JSLEE for voice and Jain Call
control. It implements Parlay to access
a media server. The experience has been
very positive for them. They have built
services using Java programmers with no telephony experience.
Question
(Comverse person) -- What have you implemented in the Parlay OSA authentication
framework. Answer – they haven't
implemented all of the Parlay/OSA Framework yet. He didn’t have the technical details on Authentication.
Clarification
from Doug Tait (Sun/Jain) – he explained how SLEEs relate to J2EE -- a
component framework, which was a bit confused in the question.
Alon – The
Parlay client API is more suited towards data services (e.g. 3G data). The SLEE approach is more suited towards
hosted and carrier communication focused services.
He gave a
lot of info on the Java community process and Java community members. It is controlled by 20 companies (10 looking
at enterprise, 10 looking at Java Micro Edition). JAIN SLEE has just gone to public review (yes, it's available on
their web site). JAIN SIP, IMAP, TCAP and JCC are all in final release
status. One thing Doug pointed out was
that some of suns competitors (HP, IBM) are active members of JAIN community (Comment
-- Microsoft is conspicuously absent).
JAIN
Certification -- to have the implementation certified, you have to run the JAIN
certification tests and be blessed by an independent company (aptest.com) which
verifies them. This gives buyers an
assurance of compatibility, and eliminates any concerns about favoritism in
having Sun evaluate implementations from others.
Doug
talked about the next JAIN community meeting -- April 23-26 in Heidelberg,
Germany.
Question
(Tony Roug -- Intel) -- How does Jain Forum link to XML based standard
(VoiceXML, SALT, CCXML)
Doug --
XML is a great standard for the industry for interworking and data
interfaces. Jain Service Creation is
looking at two approaches, one moving based on XML technologies – the hope is
to wrap up the XML based standards in a JAIN API and and make it standard for
Java. The other half is working on Java
Beans as the base. Both are moving forward.
The result will be Java/XML mappings. Tony asked of this will be
formalized by Jain -- Answer -- JTSP (Java Technology for Service Providers) is
the group doing this. They are looking at.
Tony – Do
any of the providers have an XML based interface on their JAIN compliant
product? Answer -- at least one (not
sure who he represents) is doing VoiceXML on the side. JAIN and VoiceXML sit at the same level
driving lower level APIs.
Question
(Janet Dianda -- lucent) -- How does this relate to Parlay X? (Doug) -- Parlay X is specifically looking a
level higher than standard Parlay -- web based service interfaces, not
object/API interfaces from untrusted applications into networks. Underlying
implementation is usually JAIN, but this is at a higher level. Web services mapping into the Java world
would usually sit on J2EE, and through J2EE have an interface to JAIN/JSLEE
based servers.
Question
(Dan Davidson Ulticom) -- Have Parlay and Jain Converged? -- Doug -- still in
the process. Jain/JCC 1.0 and Parlay
2.1 were done differently. Plan is to
do JAIN 1.1/Parlay 3.0 as fully aligned. (Ulticom has done a lot of work in
this and found it wound up with a JAIN implementation that was behind
standards.) Doug's comment is that the
technical issues here are not as difficult as the "diplomacy" in
coordinating with other standards bodies.
Doug --
which came first, Parlay/JAIN (Jain by 3 months.)
This
session was a lively debate involving a number of product vendors in the
area. The question of whether hardware
or software implementations are preferred for media servers is clearly still
open and subject to lively debate. Less
debated was whether to do TDM and IP out of the same media server, but also
clearly part of the assumptions underlying some of the discussions. The session was about ½ full, maybe 30-50
people.
Some
technologies take a long time to mature -- Photography took 50 years from
concept to general use (1840-1890)
Picturephone was around in the late 1950's, will it be 2009 before we
see it? (Comment -- I've never heard a period as long as 50 years. 18 was the period given by Ritchie and
Thompson in their Turing Award paper, using
a similar set of analogies and is about the time constant I now saw --
we knew how to do VoIP on grand scale in the early 1980s. We also knew how to do customer
programmability of network services.
Both are finally becoming reality.)
Media
Servers are distinguished on:
More
factors to consider:
· Service creation. What are the distribution channels, who has them?
· Wireless versus wireline -- wireless has an advantage because speech recognition is a natural interface (Comment -- but it's harder because of quality issues with the speech channel), and the business model works better (if you charge for minutes, services add minutes so you don't need to charge explicitly for the service to score -- of course many people now effectively don't pay for minutes!)
He gave a talk on positioning media servers,
basically looking at whether hardware or software implementations work better
for various applications.
Announcements:
·
Media
Server is source or sync.
·
Low need for DSP resources
·
Network
ports are the constraint, and modest size works for everyone but giant
carriers.
·
Announcements
played more often than changed.
Software
based media servers work well for this because you don't need a lot of
processing cycles (You cache different transcodings of the announcements to
save having to reconvert in real time).
IVR
systems:
Also views
software media server works well here (Comment -- unless you do a lot of ASR) It's more easily changeable and programmable
than a hardware server, and the modest size and DSP requirements won’t exceed a
software server’s capability.
Centrex/Virtual
PBX This is really IVR Plus streaming
media (music on hold), and the need to integrate with "signaling
agent" functions. They are usually
fairly small.
Another
one he thinks is good for a software implementation (Comment -> Centrex
can be huge. But most aren't).
Conferencing/Transcoding:
This is a
good one for hardware based media servers.
Port requirements are likely to be large even without the need for
extensive DSPs to do transcoding.
Call
Centers. You can think of this
application as being IVR systems plus conferencing (user and multiple agents,
multiple users and agent, etc.) Some call
centers can be very large, 10’s of thousands of ports.
Software
is appropriate for small ones, but large call centers justify hardware based
media servers.
ATELO
produces a software media server (there is an evaluation copy on their
website). It's a replacement for CTI
cards. Nothing very surprising
here. The analysis was clearly done to
highlight the areas where software media servers apply.
Impact is
another software media server company.
It's also available for trial on their web site.
Characteristics
of hardware based servers
·
provide
a complete solution (whole package including the underlying hardware)
·
The
Hardware platform is fixed
·
Signal
processing is Fixed platform DSP based
·
Dedicated
resources
·
Simple
installation -- you get the whole thing as a platform.
·
Maintenance
and responsibility is simple -- the vendor supports the whole thing. This becomes important for a carrier
considering integrating components from small vendors.
·
Hardware
solutions are generally less cost effective, especially if you start small.
·
Expansion
steps can be troublesome (i.e. new cabinet for the next T1)
Characteristics
of software based servers:
Their product is VoiceXML scriptable and
available for trial on the web.
they make a Hardware based media server, which
is fairly newly introduced.
Media
services target the services that are making money today -> PSTN
replacement, which are the core part of any next generation offering. (Comment -- he clearly views this as the
first step to next generation. I'm not
sure. Sounds a little too much like insisting an airplane land on your driveway
because your car did).
They are
adding Streaming, Gaming(?), Video, and other services that deal with IP packet
Streams.
Real
target for them is Real carriers -- the people who need high reliability, PUC
measured services, etc. The real
problem with computer based servers is that PCs and Unix servers aren't real
time, and won't stay up continuously. (Comment
– interesting that this is still debatable with so many companies making claims
about reliability of Unix and NT/XP based servers that meet or exceed those
made for circuit switches.) They
have the ability to expand to Video and new media streams on the same box is
also a big plus.
Their
solution scales to 18K ports.
They are
protocol Agnostic -- important to be able to fit into MGCP, SIP, Megaco,
etc. (Comment -- how about ISDN? or
circuit in general)
Fits
either a central solution (big cabinet with lots of cards) or distributed one
(a 1U server)
Their
value proposition -- Being able to do core voice services is a value for every
kind of carrier (Cable, LD, CLEC, ILEC, Wireless, Hosted APS)
Michael
Keefe (Cognitronics)
Migration
is important -- have to deal with TDM. Adaptation (Protocol of the day)
Media server
is mission critical -- (it's what tells your customers that you have a network
problem)
Real time
voice construction -- have to do Text-to-speech in real time. Application server and Softswtich don't know
how to do this.
Web
Programmability, needs to be able to execute service logic (comment ->
why not just a server that implements an API for something else where the web
programming is done?)
Examples
-- TDM/IN, MGCP, SIP configurations.
Cognitronics
has lots of AIN and TDM servers in service today.
Carriers
sell applications -- that's what you need. Are they the existing applications?
Those are important, a media server must do them. Not sufficient . Today’s
stuff is already bought and paid for if you are an incumbent -- no reason to
buy a media server to do a service you always can do.
Look at
Napster A real disruption, but nothing like an existing application. He made some comment about Napster becoming
the largest exporter of information from the US at one point in its history
Media
Server brings together all kinds of carriers and all kinds of
applications. It's where MP3, Presence,
Gaming, meet Cable, Wireless, Wireline.
There are
lots of interesting media based services:
What's
needed to improve success? -- Web
development model -- reduce the cost of being wrong. Let the kid in the dorm room create the next killer application. You have to have security, but be open to
anyone building applications, like HTTP/HTML or XML
Web
content retrieval -- get at anything on the internet in a scaleable way to
enable the applications.
What are
the table stakes to play?
Today the
entire base of conferencing is only 200K ports (this is for network based
services, doesn't include "3 way in the local switch" or PBX) Have to make this much more building block
approach to enable growth via new services. Future:
What’s the
killer application? we don’t know. It may be one big one for lots of ports, or
a lot of application that add up large.
You shouldn’t have to care which way you need to scale.
What’s the
scaling of software and hardware systems?
Answer (Snowshore) Cognitronics -- power of 10 Suns, Snowshore -- power
of 200 Suns Others in the 150-175 Sun range.
The message here was if you want big scaling, you need hardware. Lots of debate on this with the vendors with
hardware solutions expressing a lot of doubts on the scalability of software
only solutions.
Question: What does 200K include -- Network only, not
Class 5 or Enterprise.
Tony Rogue
-- didn't you mean to give scale in number of "intels". (Some debate
about what number of Itanium 4's were equivalent to what number of suns.)
Question: What's the difference in power consumption
between computers runing media software and hardware. Eric -- 20K watts (computer) versus 3500 watts (hardware media
server). Real issue though is
footprint. Real estate is expensive and if you need lots of ports,
hardware is more space efficient.
Objection
(from one of the software companies) -> You can do 1M ports in a 7 foot rack
(Eric -- prove it!)
Eric --
look at the number of software companies that are gone (Firetalk, Hearme,
etc.) The hardware media server companies
are surviving because it's a better solution in the real world. (Comment -- maybe, maybe hardware
companies just have an easier time in getting investment in tough times because
"software" assets are harder to quantify to investors..)
Question
(Tony): How many software versus
Hardware engineers do you have? Answer
-- 80% software -- The issue is really that writing programs dominates the
effort to build either. The
differentiator is whether it is more efficient to process audio with a general
purpose processor or with a long word microprocessor.
Roman
Shpout -- Programming is programming, you really can do these functions
efficiently. Eric – general purpose
CPUs are great at doing databases,
packet processing, etc. The place they fall
short is stream processing of the media data.
Lots of people confuse the two by looking at packet rates, but it’s what
the media server has to do that is different.
Question: Are we comparing apples to apples? Are all these servers, for example, building
the phrases from text or are some taking phoneme by phoneme based instructions
and leaving someone else the work of handling the breakdown of the text.
Answer
(Convedia) -> Hardware servers do it either way, the media server can do
both. (Jeff Dworkin from Pulver tried
to clarify this with an application to pronounce a number of dollars in various
langauges. The answer is still either
one will work. The issue though is that
different vendors quote capacities and performance in different applications).
Cognitronics -- In their experience this is better done with the media server
itself in real time.
Eric
(Snowshore) -- is integration on one box an issue? It is for an enterprise.
It is for a vendor to do a trial with small port count. How big can you scale it though. 200 ports? 2,000 ports (not likely) 20,000 ports (out of scale). Today, big
scale needs hardware. Tomorrow who
knows, that's why we all stay in business.
This
session was packed to overflowing, probably at least 100-150 people trying to
listen for most of the session. The
overall conclusions seem to be that wireless is moving in an all IP direction,
where softswitch based MSCs will have many benefits, but that standards and
intellectual property issues are holding the industry back from full
speed. Wireless has also not been
immune to the general slowdown in the industry
Lots of
potential for softswitch to enhance mobility.
He went through how a softswitch can improve economics while roaming by
allowing you to route calls directly to the mobile rather than through the
mobile’s home service provider. GSM
Release 4 has specs to do this.
Prepaid is
a very popular wireless arrangement and works better using a softswitch,
because you don’t have to set up a call before knowing the user has funds and authorization to do
so.-- don't set the call up before you verify the account.
Number
portability – with softswitch you can route to the right place efficiently.
(Comment
-- In other words all the kinds of things that IN could do if you had all the
triggers. Ravi sees the advantage of
softswitch over IN as you don't need to have the triggers.)
(Comment
Robert worked for Lucent forward looking work at one point, and more recently
for Qwest wireless.)
Wireless
clients will be dominant in whatever nation you want to have service in. Wireless growth far outstrips landline
growth.
Wireless
has lots of advantages of greater intelligence.
Wireless
versus wireline crossover for number of subscribers will happen within the next
2 yeras. Some data (Ericsson, ITU) feel
it is happening this year.
Traffic
crossover happens in 2005/2006 (presumably lots of fixed "voice" minutes
in internet access, call centers, etc.)
Carriers
are being pressed to reduce costs, and will be forced to use VOIP.
How will
it happen (conversion to IP)
Everyone
has a choice -- evolve (Lucent, Alcatel, Nortel, Ericsson, Nokia have to
address this path) Replace (Many new players including Winphoria, Spatial, etc.
) New players will create new value. 3 to one or more differential in price
between circuit and softswitch based implementations will result.
GSM
evolution -- Release 4 (2000 issue, but 2 years later implementation.)
SIP
clients -- Will first see in things like PDAs that have lots of horsepower and
programmability to a SIP API (Windows Messenger on a wireless PDA). Advantage is that you start with a computing
platform that is open to anyone and you can get lots of new applications in
this way.
Multimedia
for SIP (2005?) This assumes that you
have a SIP client that is packet voice end to end. That's why it comes later.
Communications
is far more than Voice/POTS. Handsets
are not free, they are smart computers and expensive, subsidized by
services. Churn % of handsets that
change is 24% a year -- opportunity to get smart devices in fast. (Comment -- a good point. The problem with most telecom gear is it
turns over so slowly you can't put in new technology. Wireless is different.)
The move
to All IP. IP has won the battle.
Problem is standards (3GPP, 3GPP2, MWIF) Many fights over intellectual
property.
Why
doesn't it move faster? Infrastructure
is very expensive, and it all has to be there.
Wireless data is tough to justify because the same resources can be used
to put a whole bunch of voice channels out there that collect more revenue.
Data technology doesn't need to be fast.
IMode was 9.6Kb/s for most services.
The key is good applications, and take advantage of the fact that the
phone is a full scale computer, not just a data terminal
Why IP and
SIP – Cheaper, more vendors. The
separation of bearer and control is essential.
This is a real problem in standards, where wireless data standards have
been heavily influenced by the idea that it's for laptop interface to the
internet, and too tightly optimized.
Need to support multiple levels of QoS.
Why is all
IP important -- avoid conversions, avoid signaling interworking with the
circuit world.
Overwhelming
message -> Standards groups are inventing way too many interfaces and
acronyms for the same thing. Plea to
the vendors -> Make the standards work and make them simpler.
Direct
Connect: It's a Frame Relay based
service that relays between the handsets.
(Comment -> I met the Motorola person who built it. It's interesting. They are now looking at nationwide scaling, which means putting
an IP network to interconnect islands of Voice over Frame, and lots of delay
and connectivity issues. It is an
interesting example though of using a new service as an excuse to work out a
whole new connection model.)
Capital
Expenditure for Wireless decreasing in 2002 as well, but minutes are moving
from wireline to wireless.
Went
through the GSM architecture. He went
through network architecture to demonstrate how an IP/Softswitch based network
interfaced to the HLR allows calls to be routed efficiently to roaming
subscribers. Also allows a new concept
-- virtual mobile provider. These
carriers own an HLR and sign up the subscriber, but they don't need a
network. The calls are always made
through someone else's network as a visitor.
(Comment -> Really this suggests a split between transport
provider and service provider, and a specialization. Wouldn't it be interesting if you could have your same services
on any of the 7 transport providers in your area? Transport providers compete on cost and quality of transport,
service providers compete on the value added services. Very interesting concept).
Lots of
new services can driver growth. Mobile
Virtual Network Operator is an interesting model. SMS plus instant messaging is an example applicaiton.
Question
(Motorola) -> Aren't there already too many boxes out there and won't VoIP
make it worse? Answer -- have to make
it manageable.
This was a
plenary session, with talks by various speakers on the notion of disruptive
technology and implications for the communication industry. I missed the first talk or 2 due to other
meetings.
New
players don’t want to go head to head with telephony yet, rather find niches
where VoIP works well and change the game by succeeding there:
All of
these sneak around a direct confrontation with telephony, but have the ability
to "change the game".
He gave a
very inflamatory talk, pitching out some positions:
It’s hard
to innovate these days. Innovation
comes from skunk works. It takes a long
time. Skunk works don’t survive today's
business climate.
Read
Clayton Christiansen's article in the Harvard business review, not the
book. Article says it all. (Comment -- yes, we discovered this one
before the book made him really famous).
A big thing is that disruptive technologies don't meet the current need.
Disruptive
technology is about products, not services.
Too much infrastructure for services.
(Or at least I think this was the message).
In the 80's,
there was lots of animosity between telephony and the internet. Telephone companies and people kept trying
to regulate or squash the internet, and that provoked an anti-telephony bias
from the internet community. That
carries over into some of what you see today.
Just
connecting black telephones to PC's isn't disruptive, you have to do something
better. Build something different.
Give power
to the user (Presence is the key) End to End encryption -- the end of SPAM. (Comment
– I suppose what he means is that if your incoming communication is
authenticated and encrypted, you won’t get SPAM from unknown sources, but I
don’t buy it. As long as you have some
way someone from outside your current “community” can contact you, spammers
will find a way to use it. If you want
to end SPAM, charge the sender for email.
I had a friend who wanted to patent a “sender pays” model of email as a
SPAM solution)
Past
experiences -- video conferencing with small video, audio, and application
sharing showed that customers almost always reject the video -- not useful, and
even the audio, if they can have the shared applications. Shared applications
are what people really want.
Joe went
through the definition of a disruptive technology in Christiansen's book,
evaluating each aspect on whether or not it applied to VoIP:
So, by
this defniition VoIP isn't disruptive.
(Comment -- I would actually agree.
Nothing magic about IP, "It's the internet, Stupid" ) His view is disruption took place long ago
with the internet model of data communications. VoIP is just about proving it could work for one more
application
But, we
know VoIP is disruptive, so he proposed a new reason as to why – It lowers the
barrier to entry to new competitors. He
went through the whole arbitrage scenario and how prices for voice were driven
down. The important point here is that
it's prices that fell, not costs. Lots
of people went bust, but the price came down and nobody can raise it now. The consequences:
More
disruptions to come:
Question(Me)
-- Someone mentioned Star Trek communicators, but nobody said much about
wireless. Isn't wireless the real
disruption?
Someone
(Microsoft) gave an intersting story on Voice over Cable. He had been using it
regularly with his family, and was getting complaints about dropouts. Finally brought people in to trace it and
discovered that his cable service was apparently blocking UDP traffic for 8
seconds out of every minute. This may
be a clumsy way to disable real-time communication or just an implementation
bug.
Question: Will service providers we know today be
around in 5 years?
Rinde --
yes, but they look different. Gur -- Yes, if you think of Amazon as a service
provider.
Someone made an interesting comparison about
service provider reaction to 3rd parities from two presentations made to CTIA:
The
implication is that some understand how to empower 3rd parties, and some don't.
I came in
on this only for questions due to other meetings and preparations for my
session on softswitches. A couple of
interesting ones for this panel:
1) Are we
seeing demand for G.722 and other wideband voice coding? Answer -- not much
yet.
2) On Tuesday
we heard about 3 Billion minutes sold.
Here we are hearing about 10 million ports of G.711 voice at a dollar a
port. There are about 500,000 minutes per year, which means 10 million ports is
a capacity of 5 TRILLION minutes.
What's wrong with this picture?
Answer -- not a lot of good insight.
Part of the discussion was on the fact that many of those ports are dual
use Modem/VoiP, so a lot of those potential VOIP minutes are going into modems.
This is
one I moderated. I introduced the session with two observations. Softswitch has been in the core of most
deployments so far, and most have been about relatively dumb endpoints (black
phones or dumb PC clients). The key
issues going forward will in my view be understanding where the next set of
opportunities will be, and how will softswitches play in the real world.
This
session was in a small room, and drew an overflow crowd (literally), probably
100-150 people. We had a lively discussion and kept the room full throughout
the session.
Most of
his talk was about Class 5 and Tandem replacement. Some points included using Softswitch initially to solve
Opex/Capex problems for carriers.
Softswitch is a bridge between PSTN and IP world. Second, Softswitch deployment locally may be
justified by accelerating the take rate of services.
(Comment
-- he has some past connection to Lucent Indian Hill and told me he read a lot
of the fast packet network papers from the early 1980's.) Sonus has been around before the term
Softswitch became available. He went
through the basic decomposition of the softswitch. Then went through the layering of functions. (Much like the layered models behind
Lucent’s softswitch and the layering that has been used by many other vendors
-- Bearer, Call Control, and Services).
He went through the lower levels in great details.
One point
is that Softswitches are about interworking different models, and PSTN to IP is
only one of the interworkings. The
packet world has many interworkings needed internally. (e.g. IP to ATM, G.711
to G.723) Softswitches are also, about managing connections. Even the Connectionless model of IP has lots
of "Connection management" issues in determining sessions to allow
and how to route the resulting traffic.
He went
through different packaging strategies for the 3 layers. (Something I have seen
Sonus people do ever since the Softswitch Consortium and IETF started defining
standards which functionally separate call control and media gateway
functions. Their message is that you
need to scale call control with the endpoint capacity, and the way you do this
naturally is by putting them in the same box so you grow call controllers as
you grow ports. Not bad for voice
calls, but may not work as well if you have traffic with very different ratios
of call setup and connection needs.)
He gave
some of the same talk Lucent has given above, showing the evolution of
applications from the pre-softswitch days of internet call diversion up through
internet telephony and ITSP.
Toll/Tandem transit application evolves from simple bypass to
Toll/Tandem replacement.
He went
through the SIP model and how Softswitch plays with it. Again, SIP doesn't exactly
fit the 3 layer model. User agents
combine call processing and Media.
Softswitches can play roles either as user agents or as servers in SIP
networks. Softswitches can also provide
a firewall control proxy function that allows SIP endpoints to communicate
through a NAT or Firewall which requires the ports in the messages to be
remapped and the firewall ports to be opened.
(Comment -- Aravox showed me a demo of their implementation of this
with their firewall product. What makes
it hard is having to open the SIP messages to manipulate the port
specifications in the SDP layer in order to make the connections match up end
to end.)
On Class 5
-- Their view is that it's not clear whether there is a business case for Class
5 softswitch implementation yet. The
issues around regulatory requirements and reliabilty requirements for primary
line service are the key ones.
What is
the future:
Softswitch
will be key in next generation services, SIP and H.323 (Proxies and
controllers).
Product
pitch -- Lucent learned it's lessons with the first generation softswitch --
Customers expect traditional needs for reliability, performance, and standards
are met.
Market
drivers -- Demand for broadband from business and residential is driving more
deployment of broadband and alternative service provider models.
Multiple
business models, including Virtual providers.
TDM
"Hard" switches are "Hard" -- hard to roll out services,
hard to afford. (Comment -- also
"hardened" by 30 years of experience to achieve reliability).
Key issue
for softswitches is the ability to mix/match and select best of breed
components.
Telephony
today is a small percentage of the total traffic in the network. (Comment -- but, it's still the largest
part of revenue for most carriers).
Her pitch was that Broadband networks (doesn't matter whether it's ATM,
IP, MPLS, Frame) is much cheaper and easier to management, and carriers need to
migrate their TDM traffic now to stay cost competitive.
What is
the real risk. She put up all the
pieces of a TDM switch and said that vendors of TDM switches will ask how you
can replace all these pieces. The
message is you don't. The functions get
distributed around the broadband network, and the Softswitch only has to
replace a small part of the class 5 switch.
A lot more comes from standard components (e.g. Unix). (Comment -- a nice theory, but the
trouble in my experience is the stuff has to be more tightly coupled in order
to achieve reliability and performance together. We can do better reimplementing it today, but you can't eliminate
the essential complexity of having to deal with lots of different kinds of
network failures and administrative actions, and these things get harder, not
easier, in the IP telephony world.)
Their
product is a "Class 5 Alternative" softswitch, which just focuses on
the call control piece (OSS, SS7, IN, and many other pieces come from other
components.)
Evolution
scheme for incumbants -- Migrate the class 4 first, then class 5. Everything moves from embedded servers in
the TDM network to servers in the IT network communicating through IP, not SS7.
Key issue
for carriers is ability to integrate.
You need both IT expertise and network expertise.
Old world
-> proven reliability, but no longer adequate
New world
-> Lots of advantages (but not proven reliability, and not primary line)
Question:
Are carriers going to want interfaces to IN, or reimplementation. Answer --
both. Some need the interfaces, but if
you don't, avoiding the complexities of interfacing to IN is an advantage.
Question
--- what about the signaling gateway function, where does it go. Answer -- separate or integrated. It really depends on how you want to
integrate your network.
I came in
a little late on this due to cleaning up after my softswitch session and moving
to another part of the convention center.
The session here was about lots of services and different models for how
to create them.
Services
talk -- Pingtel
Basic
message here was that putting services in central servers is still like
mainframes. Does anyone here want to
have to ask permission to run something on their PC? The beauty of the Pingtel
phone is that the applications and the control of them is on your desktop, but
it's still administrated centrally.
He went
through their model and described their “cool applications” contests. All the information is on their web
pages. Basically there is an
opportunity for developers to download the tools and upload interesting
applications for Pingtel phones, and collect revenue if Pingtel sells
them. The site is quite
interesting. He tossed out quite a few
odd ideas, like lie detectors built into the phone, interesting new call
routing services, etc. Their basic
pitch is the way you get services is by putting out a good API and making it
possible for anyone to write them. By
putting that in the phone you don’t have to worry about some of the issues you
do in a network service. (Comment –
how long before someone creates a Pingtel Phone virus, and what do they do
about it?)
He Talked about retail services (i.e.
services for retail stores. Typically
stores have IVR, and limited phone and IT infrastructure. There is an opportunity for services to
address the needs of this segment. He
gave some interesting services for retail that included 802.11 networking and
VoIP.
Isn't
there a barrier here because of the amount of capital required for installing
VoIP initially? Answer -- VoIP sets are coming down. Someone mentioned $150 as a price per set (still more than a dumb
phone). The real savings though is
operationally. Putting features on a
switch takes $1M and years (Comment -- somewhat of a cheap shot, PBX's and
service nodes aren't that bad). The
Pingtel guy rattled off a whole bunch of interesting things one could do with a
programmable phone. Bottom line for several people was that it's a more
efficient way to deliver Voice Mail, IVR, Fax, etc. Moderator made the point that you can't sell a Media server
without a core set of current services.
(Comment -- I still don't buy this one. Napster didn't provide email
or IM).
Question. What is the real first cost difference
between a nextgen environment and a TDM/traditional deployment? Answer -- today, IP based solutions are
considerably more expensive for a first cost.
Another one (Pingtel) -- IP Voice doesn't have to be more
expensive. At VoiceCON conference a month
ago (Available on the VoiceCON site), they had a set of bids on IP PBXes that
demonstrated a reasonable system cost.
(He didn't commit to what the cost difference was). Also, don't look
just at the endpoints. We traded out
Selectric Typewriters for $4K PCs. Part
of what happens with VoIP is that you have less in the core, but more in the
terminal, so you have to look at the whole picture.
Question/Comment
-- Audience person described a deployment where a VoIP IVR/Voice mail system
wound up with an annoying 5 second beep in the speech which disrupted
conferencing. Had trouble getting the
thing fixed. Not ready for prime
time. Lots of discussion about other
technologies that reached this stage and never quite made it to prime time
(e.g. WAP).
Question
-- What will help us move VoIP up to "prime time".
SALT is
Microsoft's sponsored alternative XML based interface for voice and other
media. It is based on XML based tags for
a client that can provide voice dialog capability, but it’s designed to work
with ActiveX and other Microsoft (and other’s) web technologies. It is VoiceXML like but claimed to be more
complete and more based on a web application model than a telephony model.
Paricipants
included Microsoft, Cisco, Intel.
Goal is
having a best in class way of building multi-modal services. Application runs on a web/application server
implementing SALT interfaces. Client
needs a multi-modal SALT browser that will work with your browser and interact
with the SALT service via an extended markup.
Microsoft
is very interested in having people develop SALT clients for various devices
(mentioned Symbian, Palm, did not mention Netscape.)
The talks
here were not where the action was, which was in the question and answer
period, which got quite heated as a debate between fans of VoiceXML and SALT.
Question: Where does SALT Forum stand versus
VoiceXML. Answer -- VoiceXML isn't a
standard, Standard is at least a year away.
Both build on the same W3C and IETF standards for markups and
scripting. What they think is that the
result will eventually be a standard incorporating both the dialog features of
VoiceXML but some of the more advanced XML and scripting capabilities of SALT
that result from working with a newer evolution of XML.
Question: If an enterprise has a proprietary script
and wants to migrate to something open, what do they want to do? Answer -> first decide whether you really
want to rewrite an existing script.
Moving forward for new standard, you want to get the latest standard and
technology. They feel they are building
an industry ecosystem around SALT that will give an advantage going
forward. As to whether to use VoiceXML
or SALT -- it's up to you. If you are comfortable in a more procedural model
for Telephony applications, use VoiceXML, if you are web oriented and into
objects, events, and scripting, SALT is a better way to go. (Comment ->
Interesting to see VoiceXML being painted as "procedural and Telephony
oriented). If you have to buy
something today, you can buy a VoiceXML system today. You can't buy SALT yet.
Down the road you will be a decision point where both choices are
available.
Question: Do you envision people writing in SALT or it
being generated. Answer -- both.
Question: Where do I go to learn more?
Question: What's the experinece with wireless
SALT. Answer -- none, you can't buy it
yet. Several companies in wireless
applications are members of the SALT forum.
Question: Will Microsoft support VoiceXML in their
.NET Voice SDK? Answer. Probably not. (Comment – this is not unexpected, since it seems to be the
Microsoft way, but it’s very frustrating and counter productive for the
industry). Microsoft evaluated it
heavily for about a year. They couldn't
figure out how to do real multi-modal interfaces with VoiceXML. They came up with SALT in reaction to the
problem. The reaction was to create the
SALT forum and partner with a set of companies to provide it. (This provoked a couple of people in the
audience to express their disappointment.
Microsoft and the others kept harping on the fact that VoiceXML isn't
standard.) View is that 80-90% of a
VoiceXML application, basically the application logic and grammars, would be
portable whether the presentation language is SALT or VoiceXML. (Comment -- Probably true, but not the
point. One problem with both is that
there is a lot of difficult work in getting good grammars and handling all the
exceptions in the service logic, and VoiceXML doesn't help much ith
it. I don't think that SALT will do
this either.)
Question: Relate the SALT telephony call control or
session layer to SIP call control.
ANswer -- SALT is independent of what the call control protocol and
model is.
Question: Why isn't Nuance here? Answer -- ask Nuance.
VoIP
minutes grow to 50% of the network by 2006 (IDC)
Global
Circuit Switched voice services for corporations are being retired
(Gartner). Issue is reliability. September 11th caused massive overload and
blocking in the PSTN. Didn't kill the
IP network. IP is more robust to
disaster scenarios, if it achieves promise of security, reliability, etc.
VoIP
evolution:
Factors
for success in VoIP:
Traffic
volume in IP -- 2 Petabytes a day.
Straight growth.
Requirements
for AT&T in VoIP technology
Their core
-- MPLS enabled IP on the core.
AT&T
announced an access agnostic VoIP service -- enter any way you like and hop off
any way you like.
Support
for IP endpoints is important (Metcalfe's law). When they converted Excite@home
to AT&T's network they gained LOTS of traffic. 29% of the traffic is peer to peer. (Comment -- sharing files and MP3 generates lots of bytes, but
not necessarily lots of transactions.)
Today, 99%
of the traffic is originating TDM and terminating TDM. It will be a LONG time
before most traffic is IP to IP (if 25% of the endpoints are IP, only 6% of the
calls are all IP. (Comment -> The
chart was much more interesting. It had
4 quadrants, and the IP to IP and PSTN to PSTN parts were both service
rich. The two where one end was IP and
one end was PSTN were feature poor. It
showed using “Intelligent network” to bridge services from the PSTN into those
2 quadrants. Since he never talked to
this part of it I don’t know if that implies some strategy of interfacing to
circuit IN or something completely different).
VoIP is
growing rapidly (We are in the farming business, we have planted lots of cable
in the ground. Hopefully something will
grow. It's still winter, but there are
signs of spring.)
Market is
being segmented (lots of different niche players)
Voice is
still the killer application, but there is an opportunity for significant cost
reduction.
Lots of
opportunities -- it's all about business models. Greenfield models,
deregulation models. International
models. Almost every carrier out there
is starting to use VoIP. Big
probability that you have used VoIP without knowing it.
For
vendors. It's a specialized solution
story. Protocols for 100+ countries,
including analog ISDN, SS7. Lots of
services. That’s Vocaltech’s advantage.
Lots of carriers
are doubling traffic every year (ITXC).
National
Long distance opportunities
Did a
deployment for a competitive long distance company (COTAS-Teledata) in bolivia.
Competitive carrier that provides long distance and international. Had to provide feature parity with the
incumbant. Vocaltech provided a Packet Tandem solution with SS7 gateways for
this.
Result is
a significant (30-70%) reduction in capital, and a much shorter (8 week)
deployment. They got 36% market share
in some areas within 8 weeks with quality exceeding that of competitors They have Lower Post Dial Delay, Higher call
quality. Bolivia has no pre-selection
of carriers so customers pick on every call. Holding 36% is impressive.
Hosted
Voice VPN Application.
One
implementation is enterprise builds their own VPN. Their model is like the
classic Voice VPN. The business model
is selling more data capacity. Voice
traffic on the network is almost free.
Enterprise wins by being able to outsource a lot of things they don't
really want to do. New revenue model for carrier and good for carrier
retention. Moving data to a new vendor
is easy. Put the phone on it and you
have a lot better hold over your customers.
All worlds
need to interoperate. Have moved from
an H.323 world to one where you have MGCP, H.323, and SIP together. None of the vendors today provide a complete
solution for VoIP. You have to work in
an Ecosystem group of companies that can integrate with partners to deliver a
solution.
Vocaltech
-- 20% market share of phone-to-phone international long distance. (Comment – I think he meant of ALL phone
to phone long distance internationally, not just VoIP, but I saw many different
claims around the proportion of international long distance going VoIP)
The Hype
is over, but the real work is still ahead.
Everyone now views everything wants to be digital.
Carriers
must reinvent themselves -- it's a threat, and an opportunity. VON community
thinks we are past that. It's still
percieved as risk by key carrier decision makers.
Question: Is the reason for success that Vocaltech are
the people who know how to work in 130 countries? Answer -- VOIP is the way to implement. Always involves mix of local and and vendor talent.
What's
important about the Bolivia experience is it's the first time a national long
distance competitor was built quickly using VoIP. The network uses ITXC to carry the traffic.
(He showed
a video with testimonials to them from all the COTAS management. Interesting
video, but also interesting that the delivery was in places jerky. I'm not sure what the playback technology
was.)
COTAS was
under significant attack for the choice of VoIP. Regulations basically
prevented them from doing any testing before deregulation, then originally
called for measuring quality immediately.
They talked the government into
a 30 day window of service before measuring. Vocaltech and ITXC spent a lot of
effort in those 30 days. After 30 days,
the measurement was that they were better than the incumbent.)
In the
past VoIP has been about lower cost opportunistically. Here, it was a key
enabler to do something. They couldn't
have built the network without VoIP for the same market. (Comment -- he gave market share figure
for one market, Santa Cruz. I wonder
whether this is representative of the whole company?)
The telecom
industry is not like .com. The panic is
now as overdone as the past euphoria.
Revenues are growing 10% annually (From $1 Trillion!) Problem in telecom is a four letter word --
DEBT -- too much borrowing to build out networks that were forecast based on
50% market shares.
We are in
a transition from a vertical industry to a horizontal industry. This has a lot
of pain. Will wind up with a much
better layering like the PC industry.
Growth in
teledensity is happening in the developing world. That's where the opportunity is.
India -- teledensity of 3%, but 30M lines already. HUGE opportunity as it grows.
Accounting
rate system is collapsing, but now we are near cost and they are leveling
out. Rates are asymptotically approaching
the real cost. The problem is excess debt.
All voice
moving to the internet. 6-7% today, but
a disproportionate share of the new areas where there is growth is going to
VoIP.
Non-vertical
industries need a wholesale layer and wholesale model. Retailers sell to customers but need the
wholesaler to provide them with networks and servers to back it up.
Real
economics are with Voice on the Internet, not just VoIP. The reason that ITXC is successful and
international LD rates have fallen is because it shares the Internet. Building your own IP isn't enough. (Comment – this is VERY interesting. It’s true that internet backbone
connectivity is about the cheapest and most flexible bandwidth you can buy, but
in most of my work in VoIP the assumption has been that carriers need private,
managed IP networks to get the quality.
The view you can do it with the public internet is consistent with
measurements that suggest the internet backbone actually does quite well in
delivering traffic with low, consistent delay and loss rates.)
Example: He showed a network connecting 6 cities in
as many countries with a full mesh of connections (presumably leased digital
circuits) for $380K a month. Adding a 7th city raised the price to half a
million or something like that. The
same network could be done for $45K/month (6) or $50K (7) by putting gateways
in each city with 3 times the capacity of the individual links in the mesh --
higher point to point rates supported for lots less $. (Comment -- I'm sure that the economics
still favor ITXC, but of course if you did this for real you would use a ring
or a dual hub for the private network equivalent, not a full mesh, either of
which does a better job of sharing the bandwidth than a mesh.)
ITXC is in
152 companies and has a revenue per employee of almost $1 Million a month. (Comment -- Sounds impressive, but I
don't know how to judge revenue/month for service businesses that collect a lot
of money and no doubt flow some of it through to pay for things they are
leasing.) The real economics come
because they use the internet. Of
course ITXC is not profitable yet. Last
reported quarter had $2M loss, but the loss is decreasing and he hinted at
profitability in the next quarter.
By using
the internet they avoid the need to make lots of capital investments. Thus it's a very capital efficient
business. That's good now when capital
is scarce, but capital is always scarce in developing countries.
Question: Who owns the internet you are using in
Bolivia, and how do you guarantee performance?
Answer: Much of it is owned by the incumbent. Evidence is that control of it hasn't been
an issue, but ITXC always provides backup using satellite links. (Comment -> Internet telephony through
Satellite links? That's a lot of
delay. Their customers must really be
desperate for phone service if this is acceptable).
Marconi
(Brian Rosen) (Comment -> Marconi is another fallen star of the telecom
world. During the evening I was in a discussion
on stock pricing where someone said they were selling at 5-10 cents/share.)
Infrastructure
doesn't make money, services do.
"If you build it they will come" isn't good enough any
more. The problem is that the industry
hasn't demonstrated any services people are willing to pay a lot of money for.
What's
working for people:
The big
carriers are going to be stuck on BICC (a switch to switch signaling standard
that encapsulates ISUP-like signaling over IP worked out by Lucent and Nortel)
for some time to come. SIP is really it
for the rest, who don't have the embedded investment.
Big
advantage of SIP is the distribution of functions. By allowing you to put call processing in end points and
distribute in various ways via servers you get something that scales and works.
Revenue
trumps standards -- someone who gets a service that shows real revenue that's
big enough doesn't have to be standard, but it has to be REALLY big.
All new
work is being done on SIP. Don't start
anything new on anything else.
SIP can
scale to 10-20 million endpoint networks with a modest number of control
computers (proxies). The key is
distribution. Other models with more centralized intelligence choke on scaling. (Comment – 10-20 million doesn’t sound
all that impressive to me. I see no
reason why SIP doesn’t scale much bigger.
Perhaps the comparison is to networks which try to do everything in one
central server, which would have trouble getting even this big.)
Hybrid
SIP/ISUP solutions are there, and can be used to provide SIP services for
"black telephones". (Comment – this is like the lucent PacketIN
story and some of the other application servers which use SIP, JAIN, and other
internet technologies internally but still speak IN and PRI to deliver what
they can to plain phones. It’s doable,
if you can get the right interfaces into the PSTN carrier to do it or are
willing to live with inefficient and fragile access, like routing all your
calls to an internet controlled server.)
QoS
matters. Don't use the excuse that
Wireless has taught people to pay for lousy QoS. They do it because it's wireless. If you aren't wireless, you need to be high quality. (Comment -- I'm not sure. I personally think we will see the end of
"3Khz audio" as the standard and instead have two standards,
intelligible speech (wireless and cheap VoIP), and quality speech (7Khz or
better) for premium services where it matters.)
Over
engineering to get QoS works in the core, not at the edges. Every VoIP device
should mark packets for 801.p/diffserv.
Not hard to do, and by doing it now you get it over with. Making diffserv really work though is a real
challenge. (Comment -- I believe he
claimed nobody does it, which others later refuted.)
If you
really want voice with QoS -- use ATM!?!
(Comment -- about the only person at the conference who made a case
for VoATM, at least that I heard).
With ATM we have solutions that will do 4M busy hour call attempts,
reseved bandwidth, controlled jitter. (Comment
-- Not sure if he means AAL1, which is really circuit switching in disguise, or
one of the ATM modes that's more data friendly.)
Applications. Can't be a little idea. Has to be something people will really see
as different and significant. (Comment,
but of course he said that SMS started as a little idea. So did 800 service).
What makes
a good application?
Brian
claims to be working on something like this.
We will see.
He talked
a lot about history and evolution.
Avaya demoed an IP call center to the first VON 5-1/2 years ago. Since then they have learned a lot about
what it really takes to build one.
Some
predictions from 5 years ago:
Back to
call centers. People claim you should
put the speech interface in the phone.
That won't work because the speech is context sensitive and the context
is the user's history in the transaction, which is in the call center
operator's back office, not the phone.
Where do
people call in from?
This
basically says that solutions that exploit the user's office resources aren't
all bad. Don't assume all your users
are on cell phones.
The key to
call center success is really communication enabling real business
applications. Not building the call
center as a function of it's own. What
you are building is a portal that will connect to communication enabled
applications provided by the enterprises systems integrators.
Finally,
the enterprises buy their solutions from a systems integrator, so that's your
customer (Comment -- he didn't say this in exactly these words, but the
statement is consistent with a realization I had trying to sell VoiceXML based
hosted IVR to carriers. Carriers don't
sell to businesses in many cases at least.
The business buys from a systems integrator, so they have to sell to the
systems integrator, who may be the one in a better position to make use of technology
like VoiceXML).
(He gave
his title as Disruptive Technologist.
It's interesting that this is the only place in his talk or on his
slides where the word "Nortel" appeared.
He gave an
unusual presentation using slides from a vacation.
Lighthouses
-- something that provide Presence and location. Lighthouses exploited lots of
technologies in improving what they did, but none were built for them, which is
the way innovation usually happens.
He
described an interesting service -- Caller I See -- Get a screen popup with a
picture of the person who is calling you.
(Comment -- there were bunches of hacks to do this for Unix mail and
Unix based IM systems. Some of these go
back a long way. It's cute. I really wonder about the dynamics of this
in a carrier network though. We already
have telemarketers using calling name to announce special offers and entice
people to call back. The opportunities
for abuse in graphical caller ID are all too obvious.)
Who is
your service provider?
He talked
about different kinds. Transient
service providers are one kind. That's
hotels, coffee shops and other places that give you access. That's convenient, but you need a long term
provider to host your identify to the world.
Hosting your identity means helping find you and providing information
for and about you when you aren't connected.
The classical service providers (telcos, ISPs) are good candidates for
this.
What are
some "Purple minutes services".
He showed an example using a wireless PDA to receive notification of
incoming emails, pages, and phone calls to extensions of interest. (Comment -- This is exactly the service I
kept trying to push as internet call
management. Not surprising, John has described services like this in every
conference I've seen him. I asked him
afterwards how you solve what I think is the key problem: How do you get notifications and better yet
control from lots of different places -- wireless systems, PBXs, circuit
networks, etc. He basically said Nortel
made all that stuff and could provide it.
True, but in my experience that doesn't make it work for a someone
struggling with getting the information out of 3 different carriers who are in
turn dealing with 2-3 vendors each.)
He gave a
lot of the history of BBN and the Internet.
He says that Genuity gets $1B in revenue from the internet, mainly providing
ISP service for business. He had some
interesting early history slides, including one showing ethernet running over
various forms of wiring, including "barbed wire".
Convergence
at the home -> Replacing cable, multiple phone lines with a single IP pipe
to a home hub, probably running some kind of wireless interface. (Comment -- Video over 802.11? Hard to
believe this one will really work large scale, but I agree for everything but
video).
VoIP --
needs the QoS. He had a dramatic
demonstration audio tape which gave a VoIP call mixed with heavy traffic with
and without some kind of QoS mechanism for voice. They use Diffserv and have been doing it successfully for several
years. The problem he agrees is in
access, not backbone.
Meetings
over the internet -> He showed several slides of a meeting service. (Comment -> This was basically like
Netmeeting, with maybe a bit better user interface. Usable for a workgroup with established relationships. For an alternate view on a meeting service
that might allow you to hold a "virtual VON", see
http://home.att.net/~wamontgomery/communications/conference.doc
What has
happened to our view of next generation applications?
Requirements: Core products need to be just as good as a
switch. 99.999% availability, NEBS, scaleable, fully redundant. Back office integration with existing
systems.
Need to
have a multi-vendor and mult-carrier solution for enforcing service layer
agreements. Can't look at the IP layer
as a big blue cloud any more. Have to
deal with this at call admission time.
Products
for NGN are now better segmented than before.
Products for NGN are much more mature.
Common set of standards (MGCP, SIP, etc.)
Sonexis is
a company with a conferencing product.
It came about from a collaboration with Jeff Pulver’s help.
Powerpoint
has 26 pages of clip art for meetings -- all face to face. Face to face meetings are important and we
know how to do them well. The problem
is distance. There have been lots of
attempts to use technology to support meetings. Part of the problem is that
Face to Face is Rich media.
Replacing it will require rich media.
Why
bother? Travel is hard and
expensive. Real problem is opportunity
cost. Time spent traveling isn't
productive.
Collaboration
on the net isn't like face to face and won't be like it even after lots of
works, each has advantages.
VoIP may
be stalled at the carrier level, but healthy at the Enterprise level (and thus
can be a platform for conferencing solutions targeted at enterprise use).
Their
product is designed to be IP first, with a PSTN interface on it so it can live
in the PSTN world. They believe that
the enterprise will be VoIP within 5 years.
What do
people do in meetings – He gave a long list of things none of which was
remarkable.
What do
you need:
The
experience can be better than being there.
(Comment – I firmly believe this one.
Aside from the travel related issues, “being there” means being subject
to physical constraints like who sits next to who. A conferencing system doesn’t have to be limited by these.)
Problems:
Thinks
this will become a CPE business. People
don't like to buy things "by the drink". Security will be a problem with any centralized (hosted) solution. There are counter arguments (expense, maintenance,
peak loads)
Service
providers will be needed and used for
the very large events, but every day will be done by enterprise solutions.
What do
you have to do to succeed: Think
"circuit city" it's an
appliance. Easily consumer installable
and diagnosable. Sole purpose box, NOT
a CDROM for an unknown PC.
How will
it be sold: Similar to PBXs, through
channels. (Comment -- that's not a
circuit city kind of model).
What do
you have to do to be successful?
Provide margin and profit for the channel. Easy to sell, demo, and install.
(No service calls).
Question
(Joe Rinde) - Voice sticks around too
if it gets recorded. Do the dynamics
change? Joe's experience is yes. Answer
-- don't record if you don't want to risk this.
Question
(me) -- could you have a virtual VON? Answer
– the technology isn’t there yet. Human interactions are still valuable and we
can't handle the subtleties. (Comment
– sure, but we can sure do better than what passes for virtual conferences on
the net these days, which are little more than audio broadcasts with one way
distribution of visuals)
Question -- okay what's really missing? (Handshake, signatures, and other things
about “being there”).
Carl Ford
intorduced him with stories of the last SIP summit, which started on September
11th. Carl arranged vans for people who
needed to get home and guaranteed everyone would get home.
Doug
played the JAIN video. (I don't know
whether this is still the same one I saw last time, but Shehryar Qutub and Eric
Sumner still figure prominently)
Sun -- the
number one provider of products, technologies, and services for enabling the
net economy.
Hallmarks of computing: Computing theory really dates from 1830 (bool, babbage, lovelace). 100 years passed before there was a computer.
Why Java –
It’s ubiquitous, anywhere any device. He
wished it was true for Fortran, C++, etc.
3 Million
java programmers, Java is the number one language by this metric.
Doug has
yet to receive or hear of a Java based Virus.
(Comment -- that sounds like a dare to me.)
Java
environments -- J2ME with SIP – Instead of “dialtone”, you talk to a local SIP
client in the phone, which negotiates with the network..
Lots of
Java applications, one thing that's missing is big suite of communications
applications.
Communications
is shifting -- Instead of building a communication platform and web enabling
it, take a web platform and communication enable it. (Comment -- a familiar theme if you can make it work).
Java
Community process -- lets the industry own the evolution of Java. (Comment -- JCP is really an interesting
middle ground between the Microsoft approach "we control it and put it
out, you take it", and the open source approach "Nobody owns it, if
you want it, you do it and figure out how to make it work with others")
The other
culture that's relevant here is network:
First real
general purpose computers in the network were SCP/NCP in 1980s.
The
culture class of network versus computer is evident in the API vs protocol
wars. Network people think in terms of
protocols, Computer people think algorithms and programs (aka APIs). Concern about SIP is that it's really just
ISUP with general purpose computers and ascii text endocing. The key for him is software portability and
the ability to mix and match different software features. You get this with APIs. (Comment -- If I'm a network operator
looking for the freedom to pick anyone's box and hook it up, I want Protocol
interoperability. If I'm a programmer
looking to build a general software package or an integrator looking to
integrate multiple features in one box I want API interoperability. In short, you need both).
Doug
talked about some new bets the industry and SUN in particular is making:
He talked
about 3 waves of computing. The first scaled
up to 108 devices was about computers. The second scaling up to 1011 is where we are now and
is about things that embed computers.
The third going to 1014 will be about just plain things. (i.e. the computers will be ubiquitous and
invisible.)
There are
3 places for Java in networks. Java in
the server backend is served by J2EE.
Java in end devices is J2ME.
What is missing is Java in the network itself, which is what Java for
Service providers is aimed at.
Question: What about .Net, what's do you see as the
comparison or relative roles? Answer --
It's about open platforms and open standards.
This won't be dictated by a single vendor. Java can and will ride on top of .Net, providing open platforms
and standards.
Question
(Carl Ford) to the audience -> What's the biggest company in Java
development -- IBM. His suggestion and
that of others is to look at the Linux (and open source in general) community
for some surprising indications of who is doing what in networks.
Snowshore
is a 2 year old startup building media servers. Their premier "carrier class" product is the N20. He talked about how his career and VON ran
in parallel phases:
When he
talked about how this was really playing out, it became clear that the phases
weren't sequential in time but significantly overlapped. VON is now everywhere, but we aren't done
with the transport phase yet.
You could
see the change in vendors between phase 1 and phase 2 on the exhibits
floor. Applications and Platforms
dominate instead of just gateways and infrastructure (Comment -- Yes, I can
see this but it's a gradual transition, not a flash cut. It is interesting though that in this VON
Microsoft probably had the largest booth, and probably more than all the
players who are dominantly about transport combined.)
Some
concerns for phase 2: Are we just
rebuilding Intelligent Network? (Comment
-- see Doug Tait's comments about SIP and ISUP) It is a risk, Carriers know it.
The move to commercial technology though has value independent of what
we are building. It enables new
applications, and the competition is now about new sources of revenue, not
strictly cost reductions.
Media
Content == Applications beyond the simple PSTN replacement that involve
media. Phase 3 applications are about
the web model -- content delivery plus interesting management of data and user
interactions.
The killer
"application" of phase 1 was cost reduction (arbitrage or real) – how to get a voice channel from here to there
with less cost.
The killer
application of telephony (phase 2) was the phone call. Everything is about how
to make phone calls better.
The killer
application about phase 3 is access to content. That's what drives the web.
Comment -- an interesting characterization of differences. I think this does capture something about the evolution of the industry.
Question
-- do you care about Java vs .Net?
Answer -- No, that's below where we operate, but it's a war, and wars
have casualties and cause uncertainty in buyers, so they are bad for business. Solve it and let's move on\.
This was
an interesting talk. I expected Intel
to have someone from the Dialogics part of the company talk given that that was
who was in their booth, but instead he talked about fundamental trends in
computing and wireless communication.
He started
by talking about how this was the worst recession for semiconductors yet. He had one chart that showed changes in
volume over the past 30 years, highlighting past recessions and rebounds, and
yes, in terms of percentage drops this is the biggest by far. He quoted some
business week statistics about the top 50 companies (market capitalization I
think). In 1999, over 50% were
technology companies. In 2000,
30%. In 2001, only 3, and they weren't
any you would ordinarily think about.
(EDS and a couple of others who are only peripherally about technology).
So, why is
he so happy in the middle of this?
He went
through a standard growth of industry chart showing evolution from an enabling
technology, to "irrational exuberance" with big growth in stock
prices and hopes, to a collapse and turbulence stage (where we are now), and
finally sustained growth. He talked about this in several technologies, and how
the sustained growth is always the best part of the cycle, and still lies ahead
of us. The internet is the driving
force for innovation, it is still there.
He gave
some statistics on penetration of PCs, internet, and phones in the world. The opportunity for growth is clearly in the
"rest of the world" (i.e. outside the US and Europe). Some really startling statistics. The number of PCs per thousand people in
Brazil is 36. In India it's 3. In China it's 12. 70% of Intel's business is outside the US
He claimed
that in spite of all the gloom about running out of Moore's law, Moore's law is
still good until at least 2020-2025, and Intel knows how to do it best. He had some pictures with views of device
technology in various years. In 2001,
70nm transistors operate at 1 Thz. They
have the plans and ability to scale down to 15nm. The key technologies are High K gate dialectric, a fully depleted
channel, and a raised source and drain design.
(no details presented here)
What he
talked most about was an initiative of theirs called "radio free
intel". The notion here was to
make silicon radios so cheap that they are put everywhere, virtually in every
chip. The 3 key technologies there are
MEMS (micro Electro Mechanical systems), Intelligent Roaming (seamless
migration from wired to wireless), and Dynamic reconfiguration (of the radio).
Today you
can put something like an 802.11 radio on a small card with 3 devices. The problem is it's a different card and
devices for each application. The
volume of 802.11 cards might be 100 million – nice, but not everything.
Tomorrow,
if silicon radios allow the same device to operate 3G radio, wireless PDAs,
etc. A volume of billions is possible and costs will come down.
The vision
is that the radio becomes a very small part of a chip, a design unit than can
be put everywhere, so every watch has one (no need to ever set a watch
again).
Another
trend he talked briefly about was silicon photonics. The idea here was silicon tunable filters that allow you to
create very narrow spectrum (wavelength range) optical signals that can be
densly multiplexed, as well as gear to separate and switch them. The filter is
actually a tunable reflector, which has the advantage that instead of dissipating
most of the laser energy as heat in the device it simply reflects the portion
that it selects and allows the rest to pass.
(Comment -> There are still significant power management problems
in scaling this kind of technology).
He talked
about broadband penetration in various countries. The US is in the middle of the pack in the developed world, both
now and what's projected in 2005. The
unexpected (to me at least) leader is Korea -- 35% now and 80% by 2005.
He quoted
a Forrester study saying that "real" broadband (100Mb/s) could add
$400 Billion to the US economy (Comment -- impressive, but sounds low to
me. It's impossible to predict the
effects of ubiquitous broadband connectivity). One thing he mentioned was the various projects to enlist lots of
idle computers to solve very hard problems, and how this would really be easier
with true broadband.
He talked
about the "high tech broadband coalition", which has set as a
national goal that by 2010 100M US homes should have 100Mb/s connectivity. (Comment -> I've heard from several people
now about efforts to push for a national project of this sort. I expect it may take some kind of national
effort to make this really come to path.
Even where DSL and Cable are available, the proportion of homes willing
to pay $50/month for them is generally fairly low. Just not enough percieved
benefit yet for the average consumer.
We are either going to have to find more compelling applications
($50/month buys a lot of movie rentals, so don't talk about video on demand),
or get the cost down by at least a factor of 2-3.)
Question
(Carl Ford) What will Peer to Peer services do for this? Answer -- the peer to
peer he was thinking about was refrigerator to grocery store or stove -- the
networked house. (Comment -> but
that kind of networking isn't about broadband, it's about "always
on".)
Question
What's the impact of the push for thin clients (not sure exactly what he is
really referring to). Answer -- it
helps radio free intel. Thin clients
need wireless connectivity.
What's
this going to do to capital needs in the semiconductor industry. Answer -- the only way out of this mess is
to invest. Intel spent $20B over the last 2 years on new fabs and other
technology. They are committed to spend
what it takes. It may mean though that
you have fewer players.
He talked
extensively about IP PBXs and IP Centrex.
The old
war for enterprise communication was PBX versus centrex. The first 30 years
(1940-1970) were driven by carriers making choices. No competition and slow progress. The carter phone decision changed that by opening new
channels. The PBX won over the next 20
years. It won for reasons of
technology, marketing, and regulation.
He gave
some data on PBX and centrex in the 1990s.
Interestingly enough I didn't see this as a landslide. From a 65/35 PBX to Centrex split it went to
80/20, and was very uneven, clearly driven a lot by external factors.
The new
battle in the post IP world is IP PBX versus IP Centrex. In 2001, IP PBX was 5%
of the market. In 2002, it's clear IP
is the future.
On the
centrex side there was a big boom and crash in the offerings of CLECS which
created some excitement. IP Centrex is
starting now, too soon to see it in the data.
He
described a BCR study of the University of Arkansas. They were spending $500K/month on centrex telecom services. After moving to an IP PBX, they reduced that
to $6K/month for a few ISDN connections to the outside. The capital expenditure to go IP was $4M, but
had other benefits. The result is the
service provider loses 99% of their service revenue. Not good at a time when they are all looking to grow it. The service provider gets constrained into a
commodity connectivity market, and they can't upsell services because the
services don't come from their box.
IP Centrex
is the weapon service providers must use to compete. The key advantage is
mobility -- the ability to move and reconfigure. (Comment -- yes, but it's harder than you think for a service
provider, as long as CALEA and E911 apply to IP centrex lines, because you have
to keep track of where the phones are physically to meet the requirements).
What are
the unique benefits of IP centrex -- Scale -- goes bigger than PBX and scales
over a broader range. E911/Calea
service (comment -- making lemonade from lemons?) Hosting, Survivability, and other advantages
of a robust network platform.
He saw an
interesting break point in the market at about 400 lines. Smaller than this and the buying decision is
about cost. Bigger and function becomes more important. He then plotted a diagram of the market in 4
quadrants, bigger and smaller than 400 lines and premises versus carrier. The basic threat to carriers is in losing
the >400 line market, but the opportunity is in picking up the part of the
<400 line market they don't have now. That's actually a bigger opportunity.
He also
had some statistics on current market that were interesting. Basically what's mainly happening isn't
centrex replacement but PBX replacement, at least in numbers. The competition is whether those replacements
will be IP PBX or IP Centrex.
.
This was a
session I moderated. We had about 30
people in a big room. Not unexpected
given it is late in the conference. We
had planned about a half hour of presentations and the rest for open questions. I had one speaker drop out at the last
minute which gave us a little extra flexibility. I believe we had a good discussion and brought out some
interesting points during the question period.
Talked
about the problem with PSTN based voice dialing -- hair pinning and the need to
release the call back to the switch to gain economic viability, but then having
no further control. With SIP, you stay
in control through the whole call.
Distributed
speech recognition -- Can't put good ASR into mobile phones. DSR is basically about distributing some of
the coding into the phone to optimize for recognition, not for human use. Then
can get much better recognition, but keep the control in the network server. (Comment – an interesting idea if you
actually have the ability to download media processing into whatever device is
doing the coding. I can certainly
believe that the low bit rate algorithms aren’t optimized for recognition.)
Lots of
bad multi-modal services – Today you break a voice connection to switch to an
SMS application Lots of problems with SMS reliability and roaming, loss of
continuity. It would be much better if
you could use ASR to handle the SMS as an aside, without giving up your
call. PSTN plus ASR in the net or ASR
in the handset isn't feasible. G.711
takes too much bandwidth, G.723 or G.729 don't recognize easily.
Another
class of services wants to have a voice dialog with out and today has to either
do it in a service node, then transfer the call back to the network and give up
control, or hairpin the call through the service node. Yes, there are release link trunking (RLT) arrangements
that make this a bit more palatable, but it’s not easy. Distributed recognition with pieces in the
phone which stayed active during the call would help the problem.
Best
approach combines VoiceXML/SALT with DSR and SIP. DSR solves the problems with tradeoff between coding for bit rate
and recognition. SIP maintains session
control. (Comment -> How do you
put the DSR coder in the phone?)
VoiceGenie
is working on multimodal (DSR), working with a lot of partners to give
multi-modal applications.
Works
today, deployments later this year or next.
Questions
mainly on the RLT problem (there apparently is a solution that allows a call to
be pulled back to the server if the server later decides it wants it) The real issue for him is the world is going
VoIP, using SIP with distributed recognition is a better place to start than
workarounds in the circuit network.
He started
by asking the audience who was using VoiceXML today. (half a dozen).
He gave a Fictional
scenario involving the need to look up a number for an incoming message with no
access to a data device. Today this is
tough without a data endpoint, but would be a good use for VoiceXML.
"Physics
of communication" – a plot of fixed versus mobile and synchronous versus
asynchronous. Voice has moved from
Fixed Synchronous to Mobile Asynchronous.
Data on the other hand has moved from Fixed and Asynchronous to Mobile
and Synchronous.
VoiceXML
facilitates multi-modal communication and can enable different kinds of communication
to work together (voice email, voice dialog with a help dialog).
Applications
-> Unified communication, Call/Contact center. Auto attendant. Voice
activated dialing, call screening, personalized services.
Example of
difficulty of people using features (transfer or conference). (Comment. just last week I had someone call my university number, and wound
up staring at my brand new Nortel P-phone without a clue as to how I could
transfer them. The phone had a great
display, but didn't provide much help for a novice user to figure out what to
do.)
Access to
information (standard Voice browser kind of application).
Accelerating
innovation. -> Opening up the
platform using open protocols enables more people to innovate.
What goes
in the platform? Basic TCP/IP, HTTP,
Telephony (SIP, H323, MGCP.
He went
through a lot of background on VoiceXML architectures, with and without
IP. Look at the universe of endpoints
out there. Lots more phones than IP
terminals. If VoiceXML opens up the
network to a large set of telephones and terminals, and create a "virtual
Circle".
Tellme is
a network application provider. They
support AT&Ts toll free directory assistance. Voice portal for AT&T and wireless and their own
1-800-555-TELL.
Tellme
terminates and originates millions of calls per week. Today most of those calls
come in via ISDN PRI. Could be VoIP
into a Voice over IP browser.
Basic
value of VoiceXML to them is one web
infrastructure supporting a broad range of clients (Web, WAP, phone).
Tellme has
a bit investment in PSTN interface, but they needed to have the VOIP interfaces
to evolve. Tellme has integrated SIP
and RTP into their platform. Sees the
value of SIP that you aren't stuck talking to the last switch in the
chain. It's a true session protocol
where you are talking to the user.
Tellme studio
is a free VoiceXML development tool. sip.studio.tellme.com is a portal to their
site from a SIP URI.
A problem
for them is that first tier PSTN carriers do not provide VoIP interfaces. Adding gateways are far more expensive per
port than traditional TDM. The solution
is interoperating with SIP based carriers.
Just starting to happen.
Interoperability
is good. They need good support for
DTMF on RTP (RFC2833). The non-Cisco
implementations are often wrong.
Really
working on getting SIP as an access technology from carriers. Need the non-trivial call flows --
re-invite, forking, sending the DTMF and voice to different endpoints, etc.
Questions.
-- Two questions on motivating carriers to offer SIP based connectivity. Difficult to motivate them.
Question
(me) "What's missing from VoiceXML
Question
-- SALT panel characterised VoiceXML as lacking in event model. Is that true? What's the convergence of VoiceXML and Salt look like?
VoiceGenie
supports both. Rising tide raises their
boat. Bring in new developers, web
developers thinking about speech is good. SALT hasn't hurt the VoiceXML
business. VoiceXML is here now.
Question: What are the issues on application
portability and distributed implementation?
W3C has
gone a long way to tighten up the spec in 2.0 (Common grammar and TTS
formats). Another interesting question
is whether you link applications by transfering a call, or by moving VoiceXML.)
Who
supports the common grammar format (Cisco yes, Tellme not sure, Voice Genie,
it's their speech partner. Nuance still
doesn't.)
SIP makes
the pass the call strategy much easier because you can pass the context of the
call.
Question: VoiceGenie and Tellme -> What's your SALT
strategy.
VoiceGenie
is a contributor to the SALT forum.
Nothing publicly announced but they intend to produce a SALT browser.
Question: What's the difference between a VoiceXML
gateway, Voice browser, and VoiceXML interpreter? Cisco uses the latter term. VoiceXML gateway may have more
functions in it (call management, web interfaces, OA&M).
Question: Sounds like you still need to think at the
level of a telephone call. Answer ->
we have spent a lot of time talking about the telephony aspects here because of
the conference audience. It's really
high level. People who are
This
session was also fairly lightly attended.
I went to gain some understanding of a technology that is clearly coming
that might look very different from ordinary telephony and computer
applications. I’m not sure I saw a lot
of unexpected things in this session, more like slow progress to implementing a
40 year old dream. The Microsoft
presentation raised considerable discussion over whether or not it was okay
that they were not at this point planning to follow the ITU standardization for
new coding technologies.
Video is
like audio except
H.323 is
the protocol of choice now, SIP is coming.
There is lots of it out there. (lots of companies have big deployments
of IP conferencing using video).
What's
coming:
Leadtech is a Taiwanese company doing video
on the net.
Video on the
net means different things:
Need
standards for interoperability. H.323
is the most widely used today. (He
described a whole stack of H.323 family protocols.) SIP is coming
A Video
phone is stand alone terminal device that integrates video and voice (he had a
picture of one that looks like an IP phone except that the screen is a video
display, not a text/graphics computer display.)
Video
phone service has a long history:
He went
through various scales of installations.
Large multi-site corporations will have some links operating over the
internet.
Video
phone is gaining momentum and driving deployment of broadband. Broadband connections are becoming more
common and this is a consumer type service.
It is not rocket science to install.
ISPs are bundling video phone with their offerings (presumably in Taiwan
or wherever their product is being sold).
They are
looking at integrating wireless networking to make their video phone mobile
Video
conferencing was introduced in 1964 worlds fair. Finally coming into its own.
(Comment – yes, that’s definitely the feeling much of this session
had)
Current
providers (presumably integrators that provide a packaged service not using IP)
change $1000/seat/month for video conferencing.
The network
equipment for IP video is inexpensive:
This
creates a huge opportunity for price reduction and market growth.
He talked
a lot about Internet2, the government funded initiative to build a higher
bandwidth internet that currently serves education and government users. He talked about how video on Internet2 was
used extensively on September 11th.
(Comment – One of many presentations that talked about the robustness
of internet technology during and after the World Trade Center Collapse)
He really
focused on video conferencing, specifically desktop video conferencing. Their mission is to make the PC the ultimate
multimedia device -- no reason to have/use anything else. (Comment – that’s one way to win the war
for being the platform for other communication devices)
Video may
be here and in the marketplace, but it's slow to ramp up. Real use of internet
conferencing is primarily limited to data collaboration using T.120. Video is hard to use and the endpoints aren’t
yet common.
Microsoft
is pushing to make this easier by improving XP support for devices. He described WHQL. This is the driver certification/signing process for XP. (Comment – I visited Microsoft after the
conference and learned some interesting things about the nature of the problems
with drivers and what they are doing about it.
The programming model is sufficiently complex that 3rd party
drivers frequently violate the rules in some way that leads to inefficiencies,
or failures. They are developing
testing and program verification technologies to make the developers job
easier, and make it easier to certify that drivers are correct.)
Microsoft
is working on improving the operating support for video. They are reducing
latency and eliminating glitches that interrupt video streams by narrowing the
times when the operating system (including drivers for other devices) can be
unresponsive to other input. They are
adding automatic rate control for streaming.
Lots of
application support (Windows Messenger).
Video is integrated into windows messenger.
What Microsoft
is investing in:
They are
working on using their RT Video mode of Windows Media. This they feel is the highest quality, very
simple compared to H.26L. Microsoft
owns the Intellectual property. (Will
play on all their platforms). The
problems are that it's not an ITU standard and needs tuning for low delay. (Comment -- of course the other problem
is I doubt the other players in this industry want to have to buy a license
from Microsoft to interwork with a PC either.)
A lot of
the questions here dealt with the decision on Codecs and the potential problems
that will create for the industry if Microsoft really goes with a proprietary
implementation.