This is
the 10th ICIN conference. ICIN is held
about every year and a half and always in Bordeaux. Attendance is down from the peak years, when
about 400 people came, but has been stable at about 220 attendees for the last
3 sessions. The conference is selective,
taking papers based on peer review of extended abstract and accepting about 60
out of 150 paper submissions. One of the
more interesting observations I heard was made during the program committee
meeting before the start of the conference.
We were discussing ways to increase participation, especially among the
innovators in internet and entertainment, when one of the long time members
observed that this is still primarily a conference for carriers and their
suppliers, and to the internet, cable, and entertainment industries, carriers
are the enemy and technologies and architectures for carriers just things that
could be used to clog up the network. I
believe there is a certain amount of truth in that and ICIN attracts a
distinctively different crowd from the Pulver
conferences. One indication of this is that the conference had two meeting
rooms of different sizes and in spite of the fact that the program committee
put sessions addressing new services and technologies in the larger room, many
of the more traditional topics of session control and service programming
attracted larger crowds in the smaller room.
With that said, it is always
interesting because many of the key people responsible for designing and
implementing services in carrier networks come and come prepared for in depth
discussions. Some of the themes of the
conference include:
Reza made
some interesting observations to open the conference:
(Comment – the last one here is
quite surprising given the number of people who have LOST jobs in the industry,
but we all have a very local view.)
With that
he introduced the 3 keynote speakers for the conference.
He
presented two views of NGN:
We have
always had a debate over the role of intelligence in the network. Even in IP there is centralized
intelligence. (DNS,
etc.) Standards bodies have a
consistent definition of NGN -- Packet transport, Mobility, end-to-end QoS. The IMS fits
this quite well
BT's view
is 21st Century network is based on IMS (MPLS core, IMS intelligence, mobility
and multiple services). They have a
somewhat radical view of replacing the PSTN with a packet service as soon as
possible. He claimed credit for being
among the first to push IMS as a convergence
architecture for everything, not just NGN. (Comment
– I don’t know about this. Lucent had
people suggesting this in 1999 or 2000).
Places where there are gaps in IMS include: Evolution of the PSTN; Interconnection
of NGNs, and the business model. (Comment
– I’d agree exactly on this set!)
He said
that there is not enough demand to sustain large carriers on only “dumb”
transport, therefore there is need for diversity and services. BT’s version of
convergence is service convergence for the customer. They believe in building them with reusable
components and providing an integrated customer experiences. The 3 drivers include speed to market, cost
reduction and improved customer service.
What’s
important about network intelligence – rapid service creation, but there is a
danger that the focus must be more than just building the real-time
service. His view – because IN didn’t
solve service deployment carriers didn’t really buy into it and allowed the
vendors to build their own version of service creation. That lead to
incompatible SCEs and limited deployment. Elements of network intelligence include
re-use, personalization (presence, profiles, locations),
flexibility for the future, real-time and non real-time capabilities. You have to separate the network from the
service. (i.e.
build the service independent of the network technology)
He
presented a BT architecture picture (standard IMS basically but has a PSTN Call
server that connects through a side door to the IMS.) He said that the gaps were:
He
concluded saying that the “Dumb” network is a falicy
– intelligence grows at the edge but also in the middle. The NGN model is a
good one. IMS is a good start, but it’s high risk that
we re-live the IN story. The problem
with hype is that when it burns out you through out everything, good and bad,
and start over with the next hype.
Rod
started his career doing circuit design at Bell Laboratories, but went on
through various positions to go into the Venture Capital area. Why are VC’s interested in Telecom? – because that’s where the money is. Where are the new opportunities? – “The second derivative” – where the acceleration of acceptance is
the greatest.
Example,
the internet:
(Comment – I think I understand the
concept, though relating it to the mathematical concept of a derivative is a
bit strained.)
What we
should be looking at is “IN 2.0”, the equivalent of Web 2.0
IN 2.0
has a lot of this, but also utilizes core network intelligence.
He presented “A Network Carrol”
Look at
the first possibility. Undreamed of bandwidth
(Comment – well, I know of lots of people
who dreamed of all the bandwidth a household could consume for every household
in the US and proved it could be done 10 years ago. Technology isn’t the limit, business is.) Everything is driven by “winner
take all” view of triple play.
His
vision is digital IP delivery replacing broadcast using an advertising driven
business model (“apparently free to the user”). Very cheap storage is a huge
piece of this. In 2006 you can buy 1G of
storage for 23 cents? “Exabyte” is soon to enter the literature. Terrabytes of storage for $200 in the home within a few years. Nothing every gets discarded. Lots of new sources of content (Everything
has a camera) (Comment – there are people who have been digitizing their lives this
way for some time and it raises lots of interesting questions about how to
manage it.)
He said
that people are mislead into thinking of broadband as
a New Trillion Dollar Industry.
He described
a great user generated video on the net “Epic 2014”, by Matt Thompson and Robin
Sloan. It visualizes the future
according to Google. Personalized news
feeds created by search robots that make the original news sources unable to
support themselves. “Mass
Personalization” makes advertising much more compelling to companies and
customers. Everything that’s bought
begins with a search and has personalized real-time inserted ads. (Comment
– the trouble here is that some of us just don’t buy enough to be of interest
to the advertisers)
TV is
free on a pointcast basis, and bankrupts the original
networks. Telephony goes for free, which
bankrupts the carriers, but they learn to operate in bankruptcy. “Mutually
Assured Destruction” for telcos and cable operators.
But there
is another possible future. Telcos and Cablecos not only
offer IPTV but get involved in Internet TV – adding intelligence in the networks
to participate in the delivery of video and payment (QoS). This could be paid by sponsors. New applications that value the connection (i.e.
provide revenue to the carrier for connecting the application to the consumer.)
“The
stupid network is a stupid Idea” Invented by someone who didn’t own the
network. (Comment – not exactly, it came
from someone who at the time worked for AT&T, but I guess the message is
more that the creators of that vision didn’t benefit financially from the
network)
IN 2.0 –
historically IN was about signaling, control, and management. IN 2.0 has to include a 3rd
dimension, “Content intelligence”. (IM,
parental control, security, Spam filtering, intelligent Content Delivery and
Management, Fixed/Mobile convergence)
He
presented fixed Mobile Convergence as a key application. His view is definitely
a multi-mode handset and other devices – Global fixed/mobile roaming
environment based on SIP.
Another
key is Intelligent content distribution – every song, artist, video, film ever
recorded on demand. Network distributes
anything over DSL/Fiber from storage in the core. Have to have intelligence to support storing
content locally. (Comment – his pictures look just like a 25 year old project done by the
Naperville Bell Labs called Wideband Local Distribution (WiLD)).
Everything
is controlled by intelligent policies that understand the consumer behavior,
what is skipped, what is viewed, etc.
Policies control both billing but importantly work with advertising
(Real time ad insertion). (Comment – Customized advertising has been
one of those things viewed as doable and desirable for a long time. I have become skeptical both whether it’s
desirable and sustainable, but again it depends on the behavior of the
consumer). Peer-to-peer networking
is part of the picture as a way of increasing performance – get it from a peer
instead of from the core, but still has the same policy controls on it.
To
support real video involves a 1000X increase in bandwidth through the access
networks if everyone has a different stream.
The solution has to include some management of this “Swarmcasting”
– discovering that the same stream is going to multiple places and could be
aggregated. (Comment – yes, but it’s got to be VERY smart multicasting to be able
to make the common pieces common while still supporting personalized Ad
insertion and individual control).
Huge
value of “Click through” (His example was that Google gets $100 for a click
through on a search for “Asbestos attorneys”.
(Another Bell Labs Alum) He was an
architect for BellCore’s SCP in 1988.
He
started by commenting on the trouble with acronyms:
Most of his
talk addressed a vision of Mobile personal Broadband Services in Korea
Korea means
“Morning Calm” – shy, low profile, but things are changing. 48.5 Million people,
very tech savvy. Huge penetration of Fixed broadband (almost 80%, first in the world). 2 Mb/s – 100Mb/s service rates are
available. 39 million
mobile subscribers – Almost everyone except infants and seniors. They are bringing seniors into the tech world
with training as companies target them as a remaining growth market. Korean mobile service still growing at 43% a
year, but it’s all about data. 25% of KT’s revenue is mobile data, growing quite rapidly. More SMS traffic (messages) than voice
(Calls) (Comment, but I’m sure that voice
dominates on bits and time) The average is 5 SMS
messages per person per day.
He talked
about a current IP TV service – T-DMB – Telecom broadcasting. The service is based on “Free” subscription
(advertising supported I think), 7 video, 13 audio and 8 data channels. It has 500K subscribers.
Cyworld –
combines “My space”, weblogs, mobile blogs. 10 times as
many postings as downloads. (Comment – that’s interesting. Almost sounds like a write only memory? Maybe he’s only counting download of
broadcast content and not download of user generated posts)
WiBro “Wireless Broadband”.
This is their brand name for the service, with sponsorship of the Korean
government. Basis is mobile WiMax technology, compliant with IEEE 802.16e. The service launches in 1 month (June
2006). It is now in the second month of
pre-commercial trial in Seoul, focused on downtown and college areas.
He showed
an interesting video of people using the service on a bus) Devices include a laptop card, mobile windows device, etc. User controlled services (people can
determine what services they get from who and what the payment terms are). They have “desktops” for PDA and laptop. “Push to All” multi-media
sessions. The user can access
T-DMB video.
They see
this as a step to “4G Service”. – One device for the whole world. (Comment – how Ironic. Is it an omen that I’m doing fine on my IP
based laptop, but my supposedly global GSM phone won’t work here?)
A Dream
without action is a Daydream, Action without a Dream is a nightmare, you need both.
Question (Olf Ollson, Ericsson) – what standards body works out the gaps
in IMS? Mick Reeve – OMA is most
effective but has to be blessed by ITU.
(Comment – have to wonder how the
others would have answered this) Followup – theme of the session was we have to move fast,
is this realistic? Answer – nobody is
moving fast enough. Model of doing the
standard first, then doing is best, but hard.
His view was CableLabs was the best at this,
the best standards body in the world.)
Hung Song – Was part of a panel on this in March with CTOs of carriers and interoperability is a great way to do
it. CableLabs
is good because they do the standards and interoperability testing both. Randall
– look at RSS – nobody in standards created it, but it’s very successful (Reeve
– yes, but it’s almost 10 years old)
Question Alan Lewis (Consultant) – Nobody
addressed regulation (e.g. political censorship and child protection) Randall – started
by defining politics (Poly – many “tics” – blood suckers). His response was more a warning to watch out
for political processes, have to prevent those with agendas from using
influence to force unsustainable models.
Question Jeff Johnson (University of Limerich)
How do you solve the problem of not having enough bandwidth to
support peaks – is the focus ever increasing bandwidth or better
management.? (answer,
both)
Question Huawei person – What really is the risk of
“Reliving IN?”
(Reeve) BT
didn’t buy SCEs from vendors because there was little
value in being able to create the logic quickly when integration took so long,
so instead they let the vendors build and integrate the services. A mistake because every one wound up unique.
Question (Person from George Mason
University) – will service providers invest given uncertainties. (Randall) – all
carriers are deciding that they must be in IPTV even if they don’t know what
that means. (Reeve) BT is going to do it but challenged the view
that advertising is the major source of revenue. Also some discussion on broadband wireless –
he said that WiMax won’t beat a good fixed access
structure if it’s there on cost.
Question (person from Denmark) – do you mean it when you say
change out the PSTN? (Reeve) – they have a pilot program now in Cardiff, and plan on
replacing the PSTN with an IP based structure within 5 years. Realistically they may run switches up to 10
years. The economics they see dictate
replacing the PSTN as fast as possible.
Question – what do telcos
do to survive – pipes only or value added. (All) – have to be part of the delivery of
services and add value to what goes over the pipes.
IMS is
about bringing IP to telecom. Telecom
has a history of modifying imported technologies and protocols to suit it’s needs as it adopts them and SIP is unfortunately no different. IMS is unfortunately being approached as
modifying SIP for telecom and as a result focus is very much on sessions and
calls. You need structure, but ideally
it’s transparent – forget about IMS, build your service and it works.
What needs
to happen is that we need to offload SIP from Telcom
prejudices. His example was SIP event
packages (Subscribe/notify) allowing many interesting things to be done that
have nothing to do with sessions, but not well recognized).
Another
concept was of the S-CSCF as an application level router connecting services in
the network with services in the devices at the application level. It allows applications to work across IMS
domains (Carrier networks or regions of service) To take full advantage of this he talked
about augmenting the ITU spec for IMS to allow routing of SIP communications by
the S-CSCF to be driven by profile information to allow user specific
configurations of applications.
Real
benefit of SIP is expanding the scope of what can be done –
His
conclusion – IMS is too telco focused – reinventing
the PSTN. The risk is that others use
the same architecture and technologies without the PSTN bias to out compete the telcos.
Ron’s talk
was basically along the lines of how we risk following the same paths as we did
with previous services architecture, while making the world more complex with
more intelligent elements to manage. One
interesting comment – IN built on SS7, and there were was a case for the
infrastructure needed independent of services, anything that IN added was
“gravy”. With IMS, the infrastructure
isn’t independently paid for and thus the business case requires that increase
in revenue or decrease in cost pay for the deployment.
(Comment – yet another alumnus of
Bell Labs.
I wonder what percentage of speakers share that and what the impact of
the demise of Bell Labs as a common heritage for the leaders in telecom has on
our industry) His role most relevant here is as chair
of the OMA work focused on mobility. His
message was on the importance of Social Software Applications as a driver of
the new communication paradigm. By that
he meant things like Skype, Google, etc. He talked about the standard operator
paradigm. You have a relationship with
your operator who knows your devices and is responsible for interworking
with others. The alternative in the
social networking applications is that the network is transparent and you
network directly with your peers. You
are more in control. Voice is an add-on
to other things the service does for you. The value is in the social networking
services.
Question – (Roberto Minerva) Is IMS a good
response to competition like Skype? Christophe – IMS is
good, 3000+ people in a telco should be able to beat
the 100 people in Skype, but our culture is pushing
us in the worng direction.
Question (Bernard Villian) – Is there a real
threat from Skype and Google or are these just
getting new subscribers willing to tolerate low quality for “free”. (i.e. would a
business use these services as a replacement for telephony?) (Comment
– nobody raised this at the time, but I firmly believe that this is NOT the
case. Skype is
gaining traction for internal communication in Personeta
because it is better than the PSTN. Broadband
voice, easy conferencing, presence status, and it does
a good job with echo management, better in many cases than a PSTN conference
bridge. Skype
has another odd benefit which is that it forces the user to use it from a PC
with broadband connectivity so on Skype you don’t get
someone calling in on a mobile in a noisy environment with no access to
reference materials)
The
Extended Presence Concept: aggregate
context information available from the user and carrier into one concept. Their EPS is a server that provides service
creation based on aggregated presence. The
server uses XML based encoding of presence information with Web Services/SOAP
interfaces. EPS has a layered structure that looks a bit like TappS. They have a
rule manager that integrates the presence attributes with the services and
controls synchronization among different services.
A lot of
what she discussed was user management of presence aggregation, respecting the
reality of being mobile and having multiple presence profiles. Idea is user signs on and adopts a profile
which then supplies the proper integration of the user’s presence and
context. (for
example, a conference protocol which provides availability only for critical
communication and limits media types as appropriate).
EPS
functions as a presence broker, interfacing to many
carrier networks and devices. She talked about the uncertain world of standards
for integrating and sharing presence and identity today.
IMS is
the most important architecture to Telcos now, but
IMS greatly increases the number of smart elements that have data. Management and provisioning of data in all
these elements is a key problem. Subscriber
data must be spread over multiple nodes.
Another new factor – end user provisioning of data.
This makes it harder to prevent errors.
Two options have been proposed:
Data
architecture is key to IMS. Database technology is becoming the incontrovertible
choice for a basis for doing this.
He started
by recalling the large number of papers on presence based services in ICIN
2003. (Comment – I think I was one of those presenting one, though that may
have been my Lucent paper in 2001 that I did not have a chance to present) Now people are building presence based
services, but there are snags.
He started
by reviewing the standards activity – in summary, too many and very
inconsistent. Problems created by need
to be open (need to be able to exchange information across different domains
and different architectures)
He talked
about a lot of the challenge is integrating presence from multiple sources and
in exchanging presence between applications
Lots of the challenges are intra-carrier – interworking
between incompatible systems in a carrier.
He
presented a concept of context based call forwarding by describing a scenario
where someone calls into a meeting, realizes that they need to have video
connectivity, so he moves to a video booth and then gets the session moved to
the booth to allow him to participate by both video and audio. The cell phone originally defines his
context, but when he shifts to the video booth he gets a new context with video
capability. (Comment – this is one we could easily do, probably do more twists on
it) He said that the automatic
context monitoring could be used to do this by changing the call based on the
user’s intent. He contrasted this with a
call center solution to the problem.
Call center session movement is by agreement of both parties, while the
context based movement impacts only the party whose context changes.
Some
limitations they found in doing this:
SIP Refer, the obvious way to do it doesn’t have a way to distinguish
call migration in the two scenarios. In
SIP you can’t invoke call changes from outside of the session (Comment – not sure why you would need to do
that). There solution – introduce a
new SIP method – CCR Request. (Comment – yes, unfortunately that’s what
everyone does in SIP – if it doesn’t do what you want introduce an extension). Lots of issues with this around trust, since
the CCR Request comes from outside the session, how do you authenticate
it? That wasn’t part of their work.
Question – Why invent a new method, use XCAP
to allow the user device to inform the service of new capabilities
required. (Answer – yes, that would work,
but the CCR is more convenient and less constrained (Comment – yes, but that’s part of the problem). Using XCAP requires the user’s terminal
understand how to generate the response.
Question – How do you handle the performance
issues created by having too many presence change notice activities? It’s not just about managing load from the
intended uses, but hackers can introduce a lot of presence change traffic with
little effort and essentially deny services.
Presence services can put the user in the position of network
administrator – user has to manage bombardment with presence change notice
traffic. Could become
exponential growth). Answer,
isn’t specific to presence. RFID data
generates the same overload threats. (Provenzale) – yes we know about
this and we don’t have a good solution.
What happens when a user moves from one cell to another or makes a minor
change in availability? Does this create
a burst in traffic for the limited radio channel to generate presence
notifications? Have to limit user’s
presence traffic.
(Sammateu) – Presence broker can help here by filtering
presence information to hide complexity from services and endpoints that don’t
want to deal with it. Have to put
intelligence in presence aggregation. (Comment
– this is not unlike the solution that Personeta had
to come up with to do converged desktop in order to separate the presence
status change traffic from the other traffic so as not to swamp call servers
with presence traffic due to registrations and changes)
Question (for Bihan)
– Slides differ from paper in having one more layer, what’s the other layer
for? (Answer – drawing was different but
idea is same – 3 layers, Application server, “real time part of data”, and
database.
Question (for Provenzale)
– why converge user’s identities, users have multiple identities each with it’s own presence rules.
(Comment – this is exactly what
was in the paper I was referring to.)
Answer – yes, that’s right, users have multiple identities and their
status depends on the user’s communication capabilities.
He talked
about calls involving a mix of circuit and packet media. “Push to show” – services which start with a
voice call and add additional media (send picture, chat, etc.)
He talked
about many services that used terminal based applets together with web services
control functions in the network to enhance circuit calls (e.g. using a network
based address space or using network based personalization of greetings (ring
tones).
Network
based vs Terminal based enhancement.
John is the
leader of HP’s IMS architecture. I know John from previous ICIN and VASA
activities. They looked at two different approaches to group communication
through an implementation experience in their lab (HP Grenoble)
– one based on the IMS standards and one based on service chaining according to
Web Services.
5 years
from now we will have new services. The
real question is whether IMS is going to help it. Why interest in SDP? IMS standards address the transport and
mobility but do not completely address the service layer. SDP proposes a way of reducing the cost of
service introduction through layering (common enablers on which applications
can be built.) SDP can be combined with IMS turning the underlying capabilities
(e.g. HSS) into re-usable components.
They
started out with some use cases for enhancing a call with rich media. They did it via a standard IMS application call
flows and reached the following conclusions:
The second
part of the experiment involved service chaining, a concept in web services for
combining service elements into a single service (Comment – it’s a lot like passing control among components in a
component environment) Service chaining allows
multiple individual services to be combined in one session. This has to work across multiple devices and
networks. Chaining can have sequential
or parallel invocation. (Comment – I don’t know if there are
standards for concurrent invocation where multiple services are active at the
same time) The SDP concept uses XML
and web services to accomplish this.
They added
a service composition element, a service registry, and a service context
repository. The registry registers
components, and the context repository holds context information passed from
one service to another.
Conclusions:
Traditionally
enterprise IT and telco networks have been separate
functions. Convergence is about converging these two.
Telcos desire to provide the converged
functions. Enterprises benefit by using
the converged network to enhance their business.
He defined
convergence as a process, not a single step.
SDP is a concept, not formalized and standardized. Service Oriented Architecture (SOA) is standardized
by W3c, but the standard is technology oriented (Comment My view is that this is more syntax than semantics. Syntax is important in being able to
communicate, but without standardizing semantics (i.e. functions) you don’t get
re-usability) Convergence process defines
reference points where interworking can occur.
SOA defines
points of integration using XML to define interfaces. Many different kinds of software use
SOA. Enterprises gain technology
independence by defining a process as an integration of components, each of
which uses web services interfaces and can have any technology inside. BPEL is a language to assemble the
components.
SDPs
implement reference points inconsistently today. Model is that services export integration
points, which are integrated into applications, and applications connect to
users. The integration of applications
and services takes place in what is known as the distribution plane.
SDP can be
defined as a distributed IT platform architecture using generic SOAs to abstract infrastructure functions and enabling 3rd
party implementation of applications that use those abstractions.
Question (Chet McQuaide)
– how does this relate to the split between SCP/IP in IN? John O’Connell – media functions are more
complex in IMS because of richness and standardization of application
environment (e.g. J2EE).
Question (Bernard Villian)
– Spotlight on Web Services. Are Web
Services more broadly applicable to our problems (example migration from legacy
to IMS), but also in addressing operations.
Answer (John O’Connell) – moving applications into web services gives
flexibility to apply them to either NGN or legacy, but not all applications
benefit from that. Some require more
performance and don’t benefit from network independence. For most though moving
applications into web services opens the environment for a greater development
community. “ISVs are not familiar
with JAIN/SIP”. They can use Parlay X or
even proprietary interfaces exported by telco
networks.
The
presentation described an IMS testbed constructed by
FOKUS and looked at the suitability of various technologies for
applications. IMS has 3 alternatives for
building services:
This paper
really focused on SIP, and specifically on a SIP server built using SIP Servlets. SIP servlets are a mechanism in Java that triggers particular
actions based on incoming SIP messages and events. XML is used to define the specific events
that a particular Servlet handles and the interface
to that servlet.
Fokus has two “Playground” testbeds
for services, one based on IMS and one on Parlay/Parlay X. Their SIP environment includes the HTTP servlet interfaces as well as Open Diameter, used to
support the interfaces for HSS. A C++ implementation
of the RO interface for charging is provided with a bridge to allow it to be
used in Java.
He
described using the SIP environment to build a service which automatically
completed a call when the user went online and became available. He also went through a simple call center
example using Multimedia.
Question
How
Portable are Sip Servlets (The person asking said his
experience was that they were not)
Answer – their implementation conforms to JSR116 (SIP Servlet) but the applications access HTTP and other
services. The part that conforms to
JSR116 is portable, but interaction with other services that may not be
available in all servers may be a problem.
(Comment – this is one of the key
portability issues that makes me skeptical of any claims about application
portability for a REAL application, it’s hard to build a complete application
without interworking it with services in the
application platform which are not standard.
The deeper problem though is likely to be with SIP itself. Each time SIP is extended or new patterns of
use are developed for new services the processing required changes, and the SIP
Servlet interface exposes those changes to the
application so that an application written for one version of SIP will likely
require modification when the protocol is modified.)
He began by
talking a little about Appium, which is an
interesting company in the industry. Appium is about 10 years old and has been building
applications for carriers independent from the major
vendors. I know Appium
has been engaged in Parlay and SCP implementations.
He talked
about different sources for Applications:
Off the shelf, Operator developed, Operator commissioned, and 3rd
party. He also talked about 3 classes of
services: Pure IMS, Convergent, “Ported
legacy”, and “Legacy SIP”. (Comment – this was one of the few talks
which acknowledged that there is and will continue to be a base of SIP based
services not built to the IMS specific versions of SIP and that interworking with these, which occur both in enterprise
products and in early carrier products, will be a requirement.)
Next he
went through the programming requirements of various kinds of services and the
capabilities of various kinds of servers.
The IMS SIP application server has to deal with several other interfaces
into the IMS: HSS, CSCF, MRF, Charging,
etc. Parlay on the other hand abstracts
all these through the Parlay framework and services APIs. “Microsoft has a SIP interface but it’s quite
limited, not recommended”.
OSA/Parlay
Parlay X –
Same basic characteristics as Parlay but more abstracted.
SIP Servlets
JAIN SIP
The
presentation showed a matrix of these models and the various classes of
services and indicated where each was best, indicating
for example that Parlay and Parlay X met the needs of 3rd party
developed services, while the JAIN SIP api may be
better for the core carrier built ones. (Comment – this is an interesting
exercise. I don’t know how universal the
criteria for categorizing different APIs are)
Question (Chet McQuaide)
– On Monday someone said IMS is a Docking station for application servers, do
you agree? (Yes)
Question (Warren Montgomery) How are Service Interactions Handled, specifically when
applications were built independently?
(Answer – you are up the creek without a paddle) Seriously, SCIM is defined as to what needs
to be done not how. Feature interactions
have been an issue for a long time and nothing has improved much since 1994
(apparently when he did a Ph.D dissertation in that
area.
Question What differentiates JAIN/SIP from
SIP Servlet?
(Answer – SIP Servlet is locked specifically
to SIP, the application can’t call other APIs.
JAIN SIP doesn’t constrain the application so it can be used by applications
that are also calling other protocol APIs.
The paper was about enhancing the way invocation servers are
invoked. Today, the S-CSCF uses InitiatialFiltering criteria(IFCs) to determine when to trigger an application
server. The problem is that the only
TIME an application server can be triggered is on the SIP Invite message. Once an AS is inserted into the call flow, it
is part of the call session until termination.
What this means is that a lot of application servers have to be put into
the call flow on the first message on the chance the feature they implement
will be used, and then stay in the call for the entire call. Consider the HOLD feature. To enable the call to handle HOLD, the AS
implementing HOLD has to be inserted as a back to back user agent at the
beginning of the call. This
substantially increases delay and processing load without doing any useful
work.
Their solution was a SIP Invocation Function (SIF).
Question
(Chet McQuaide) – Do you consider
SIF as part of the SCIM? (Yes)
Question
– IFCs are unstructured and low
level. Not as good as today’S IN (He agreed)
The speaker explained their concept of the service delivery
platform – a central core with extensions, using standards where they
exist. The platform was layered where
the core supports the key services (including SCIM), then a web services layer
connects the applications. (I could not stay
for the questions for this presentation due to the need to prepare for the
session I was Chairman for)
This is the
session I chaired, so my notes are somewhat sketchy (hard to take notes on
stage). I introduce this as a critical
and often under covered topic, and that neglect of these areas is really what
limits the deployment of new services. I
suggested that NGN in general just makes this works because there are a lot
more intelligent elements in the network requiring provisioning and management.
Claudio was
a substitute presenter but did a good job.
The basic concept was using a SIP based “Service Bus” to handle
distribution of provisioning information and collection of measurements and
other data. This was done by
Subscribe/Notify. In principle this works quite well, services discover
elements in the network through notification and get their information
well. There is an issue of how the
system scales without using some kind of multicasting, especially if it is
extended to end-user devices.
The speaker
is a Ph.D. student and presenting thesis related material. LINDA is a non-procedural distributed
programming language based on the notion of a big shared “tuple
space”, that autonomous processors can post to and can receive notifications
based on a match of criteria. Mobile
ad-hoc networks are a real challenge because there is no central authority to
build any kind of structure for service location, charging, etc. The solution using LINDA proposed having
nodes merge their tuple spaces as they intreract, allowing users seeking services to discover them
dynamically as they come in range based on notification criteria. (The real model was more complex, with
application suppliers, application servers, and capability resources used by
services all of which are needed to deliver.
The
solution seems interesting, but the merging mechanism and mechanics of scaling
aren’t clear.
This paper
presented the concept of a home gateway at the interface between the home and
the network in enhancing communications.
(Comment – nearly 30 years ago
while prototyping ISDN services we invented the concept of a home gateway,
which terminated the ISDN and provided access for a variety of devices in thehome. Our
“gateway”, was a minicomputer, but we envisioned it being shrunk and assuming a
very similar role.)
He made
some interesting points about the home environment being chaotic (multiple
devices, multiple networks, multiple identities. Some services are associated with the home,
not a specific device (i.e. call someone in the house to notify them of a
delivery or emergency). It has to be
very easy to add devices and access services and content. By putting a home gateway as a proxy between
the local environment and devices and the network you can centralize
responsibility for identity management and DRM and simplify many things going
on (Comment – you have to be careful not
to allow the gateway to block future extensions and get in the way)
Services
were composed by chaining together service elements. Services were composed through a template
which layed out how service elements would be
composed together. They did some
performance studies on how performance varied depending on the characteristics
of the Service elements and service templates and as expected, big services and
lots of services caused degredation.
Claudio was
also a substitute speaker on a paper in my session. Telco operators are interested in SOA and
service composition because they want to simplify service creation, get reuse, easily integrate with non-telco
elements. (Comment – I’ve heard this one before, which was part of his
observation as well.) So far telecom
services get delivered into separate vertically integrated service specific
networks, which limits sharing and re-use (Silos).
Services
can be categorized:
He proposed
an architecture with “content” web services and Telco
web services as elements, then composition, which build services from service
elements, and the ability to expose those composed services further.
Challenges
for mixing different service types:
He
described the BPEL4WS web services standard, as seen from a telco
perspective:
They were
very interested in JAIN SLEE (jsr240) which was developed in parallel to
J2EE and implements an event oriented model.
They have a
prototype SLEE – adaptors with an event router on top to connect it to
components in the SLEE. (Comment – like TappS
without the virtualization of the protocol interfaces) They have been
successful in using this as a web services coordinator in telecom services.
“BPEL” is a
Turing equivalent language – you can express anything if you really want, but
it depends on how much programming you want to do.
I
unfortunately came in late and missed the first paper from Telefonica
Moviles relating to virtual network operators. I know from reading the paper that it covers
how to use the IMS architecture to support VNO operations but was unable to
hear the presentation.
His paper
presented a lot of information on the expected consumer behavior with mobile
TV. The basic message was that because
it is a unicast personalized service, while the
amount of time spent watching may be lower than “broadcast” TV, the value to
the subscriber is higher. Also the potential to customize the advertising
content delivered and thus the value to the advertiser is much higher. The data given was a bit confusing though.
His paper
discussed issues with distribution of Mobile TV, including the key requirements
for the service, quality of service, coverage, and usability issues.
This was an
economic model and analysis of replacing a circuit/PSTN architecture with a VoIP network based on IMS.
The real cost in the network is in the access (Comment – no surprise here).
His model assumed a DSL based access network carrying IP voice to the
premises. The conclusions were a bit
vague because the economics are very dependent on assumptions about the access
network (usage and density) while cost savings result in the core network
(elimination of separate circuit/packet networking and reductions in
operations). Fixed/Mobile convergence is
seen as a major driver for PSTN replacement in eliminating separate networks.
He
presented some standard models for predicting subscriber behavior in choosing
new services and then actual data from Docomo’s
services to validate the models. His
predictions in general were reasonably close.
One factor
to consider is price elasticity, which he modeled saying that there was a
maximum and minimum use of a service which apply when the price is extreme
(i.e. if price gets high, there is a minimum communication required) in between
limits though users tend to consume at a rate that keeps their payment
constant.
Finally he
had a profit model which takes as input subscriber growth, price elasticity,
and pricing plan together with statistical data on the network use and the
result shows traffic and revenue growth over time with confidence limits. (Comment
– I was too far from the screen to go through all the details, but there is a
lot of detail in the paper on exactly how the uncertainty in modeling was
handled.)
The speaker
was from the Bell Labs organization in the Netherlands and described a research
prototype.
The paper
talked about a gateway for charging that interfaces with SIP sessions and with
the diameter based IMS charging interface.
It is in fact a SIP to Diameter protocol converter of sort, but has to
have session awareness. The gateway sits
in the SIP flow and generates charging information based on awareness of the
charging type (e.g. flow or event) . He went through several scenarios including
charging by reservation, where the gateway is continuously updating the
charging during a session by timers, or mapping a SIP message with specific
charging information into an equivalent method in Diameter.
Some issues
in mapping SIP to Diameter include having to understand how to map Session ID
information, and dealing with parameter differences in how the Diameter
information is encoded based on IETF standards vs the
ITU. The conclusion was that this was
quite feasible to do, and that the same techniques could be applied to any
other protocol with session information (e.g. COPS for policy session).
He talked
about the need to support new kinds of charging in order to support content
delivery. Real-time and differential
charging (based on time and discounts based on the subscriber) are important in
marketing these services.
He went
through the existing traffic based charging in GSM/GPRS networks, generally
done by the GGSN (for data).
The
proposed solution in 3GPP is IP Flow based charging. The concept is that the actually content of
the IP flow is analyzed to determine how it should be charged as it passes
into/out of the network. This allows an
operator to centralize the analysis and charging and eliminate the need for
multiple access points to distinguish charging rates. It works by introducing the analysis function
that monitors the flow at layers 3-7 (i.e. can pick out specific application
messages to determine the media types and charges), then interacts with the
charging functions in IMS to implement charging.
Some of the
issues include how to distinguish information.
For example, if you identify SIP from the ports used the user can hide
SIP by using other non-standard ports.
When SIP is identified, it may still not be possible to distinguish
among multiple uses of SIP (IM, session control, subscribe/notify, etc.)
T-Commerce
is TV based commerce using a mix of broadcast and unicast
information and a set top box to facilitate upstream interactions. Much of the talk addressed making the service
attractive to the end user. He talked a
bit about what limits consumer use of electronic commerce systems. The big issues were distrust, especially
security. He presented some information
about how to insure that upstream payment information is not mis-used. (This was
a somewhat difficult talk to follow from the back of the room)
How people
make buying decisions evolve over time:
Making the
experience good requires a lot from customer care. Sign up, billing, provisioning, discount
plans, usage reports, etc. must all fit how the customer wants to manage the
service.
One thing
he spent some time on was the whole issue of controlling usage to limit risk
both by the consumer and the carrier. For example limiting usage when a threshold is crossed to eliminate
video while continuing to allow audio or emergency services. Accounts need to be refreshable at will by
the consumer.
Traditional
billing architectures don’t handle this well and tend to be smokestack
oriented. His solution was to
restructure charging in horizontal layers, so that all services go through the
same charging structure which can apply limits and handle refresh. (Comment
– I am not a billing expert, but this didn’t seem radical to me)
Question (For Bell Labs Speaker) – how do
you get the SIP flows to the charging gateway?
Answer – Charging gateway works as an application server. (Followup – wont
other application servers be impacted by charging
actions, for example exhaustion of charging accounts should be able to shut
down some services.) (Comment – this was clearly inspired by the
last talk and points out the problems of a structure where charging is a peer
of services rather than a foundation.)
Lots of
Hype around peer-to-peer, but it’s important to separate the technology from
some of the business aspects.
What is
P-to-P – each participant takes responsibilities. New paradigm for communications (Comment – not really, consider almost any
radio service like CB radio, Ham radio, etc.) He talked about this as being good for some
things and not others and that the analogies of IT in the 1980s are good –
everyone migrated to PCs, but discovered that PCs have to be administered and
that is work.
Not a new service, P2P file sharing, Skype, etc. Siemens and Avaya have IP PBX architectures built in a P2P architecture.
This work
focused on providing multi-media over “the omnipresent internet”. He overviewed the characteristics of
centralized vs peer-to-peer. Peer-to-peer is inherently best effort, and
relies on the resources of the client peers, while the client/server (centralized)
model you have operator provided infrsastructure and QoS guarantees. Both
will scale, but in the client/server model you get escalating complexity to do
it because the central elements must be scaled, while peer-to-peer usually
scales more naturally.
He proposed
a model that captures the best of both – a peer-to-peer architecture of
operator provided nodes in the network with a client/server relationship
between the users and operator provided nodes.
(Comment – this sounds like a very
workable idea and with the right protocols and ability to change binding
between client and server according to load and failures would have the ability
to support high quality service)
He talked
about two different models:
Question (Dave Ludlam)
– you claim to be able to deliver QoS, but the
reality is that P2P networks deliver good QoS today,
so where is the value for the end user who will have to pay more to the
operator to support an operator provided solution. (Answer – things are fine today, but video
may create QoS problems. In addition P2P networks lack customer
support so for some customers having the carrier answerable for service issues
may be very important.
Question Are there any applications not suited for In network
peer-to-peer? Billing? (Answer – they haven’t found anything yet
they can’t do)
Really
applies to both Mobile and fixed. The
work was university of Helsinki work, not Nokia. One of the big motivators was to boost the
usage of mobile data. (Comment – this is interesting. North American carriers are very interested
in mobile data but I thought that in Europe and particularly in Scandanavia
would have more of it.) The answer
of course is price – if sending data on your mobile costs money and sending
data via WiFi or fixed is free.
He
presented a mobile data client that could use PC based P2P services. Mobile data is currently Interesting, but
expensive (data connectivity).
He
presented some speculations on how to use mobile capabilities – what if you
could google “ICIN 2006 real time” and get a menu of
people with mobile phones that are video capable attending ICIN to get feeds
from (Comment – yes, but the problem is
the cost of sending data)
Other
scenarios included organizing pictures, audio, etc. The big issue with mobiles is battery – you
can’t afford to share your battery by leaving your mobile on and available for
uploads all the time. Any solution for
P2P mobile must acknowledge the limited battery.
He talked
about the long tail – the ability to find rare things over P2p. Amazon gets a lot of money from books that
are rare but there are lots and lots of them
Free riding
(downloading but not sharing ) is a common
problem.
Price is a
big problem. For a song download
He said iTunes gets good use even with the price but the mobile
premium is too high. (Comment – It always
amazes me what people will pay for something in small amounts at a time. 1Euro is more than the price in the US, which
I thought was high, but I’m not the target audience for music services)
He talked
about an architecture for mobile peer to peer. This used a mobile proxy, which would
represent the mobile in the internet peer-to-peer service. It handles addressing problems (Mobile
doesn’t have a permanent IP address to be routable), but also legal issues like
blocking illegal file sharing. (His view was that putting that on mobile
devices would make the operator subject to lawsuits)
He went
back to dynamic searching to discover data available from peers, talking about
this as a real weak point in Google – the information isn’t up to date enough
to find things that are happening right now.
Again battery
is an issue and there may need to be compensation to the peer providing access
since a real resource is being depleted to send the information.
Question
(Me) – if the goal is increasing data use, isn’t the obvious problem usage
based billing? If you can get it “free”
over fixed internet and Wifi, why would you pay for
it over mobile data? (Answer – graduated
billing models with steps (flat for a while) might help, but also content available
uniquely via mobile like live videos and pictures from events are probably the
thing it takes) (Followup
– Me – yes, but will the owners of THAT content allow the use of mobile devices
to share it as acceptable use? Probably
not for commercial content (concerts, etc.), but perhaps there will be enough
interest in individual content. (Comment – 3 years ago I had the good fortune
of being at Hawaii’s Kiluea volcano during an active
eruption with surface flows. We could
hike right up to the moving hot lava, and there I found many Japanese and
Korean tourists with mobile cameras snapping and sending. Maybe there is a market like this. Who knows, maybe you can pay for your trip to
Hawaii by agreeing to share your content.
Still, I suspect that the real moneymaking uses will prove to be
X-rated.)
He talked a
lot about how to implement SIP in a fully distributed system. He started with a search algorithm, DHT (I
think distributed Hash tables), which would be implemented by multiple nodes to
support databases. A key problem is
registration and location. Using DHT a
node hashes it’s URL and registers with the node
located by the URL hash. Another node
can then retrieve that information using the same hash to locate the same node
to retrieve from. (Comment – works fine if both parties know the same set of nodes!)
He went on
to describe IMS implementation locating HSS data dynamically and triggering
services to applications. Again, it
seems clear that DHT can work for distributing responsibility, but I don’t know
how it works unless
all nodes know exactly the set of other nodes participating.
He then
went on to talk about how to map a user to a SIP User Agent using this. (Comment
– why does this seem a little like building COBOL emulation on Java? Maybe useful for some things but if you have
a working P2P network why make it emulate something more centralized?)
Question Will this work for Mobile? Answer – the problem with mobile is that it
costs money to participate in the network so there may be a problem getting
everyone to participate all the time.
Question Will data be lost if a peer
disappears? (Answer – no, there is a
replication mechanism where data is replicated in neighbors to maintain it if
one node goes down, but there is also a cost of extra “hops” and DHT may not be
the best answer. He suggested a
hierarchical peer-to-peer model which is apparently how Skype
works.)
Question (telecom Italia) – how do you do
network embedded features like lawful intercept? Answer – this is an outstanding problem. Maybe an operator needs to do this on the
access network directly
Question – Media, like Video, will probably
dominate the network bandwidth. Have you
thought about whether P2P can support caching to reduce network traffic? (Raivio) – Mobile
proxy can cache. (DeVos)
– caching is needed and can be done in local or edge nodes (his paper was the network peer-to-peer) (Followup – thought Bit-Torrent was the ideal example. Bit Torrent finds the nearest source of a
requested video and uses it and can have multiple downloads in progress and
merge the results). (Lindqvist)
– Pricing has a big impact and pricing can change rapidly.
ICIN had a poster session for the first time in recent memory. The papers accepted for poster presentation were one the ones just below making the cut for presentation in the sessions. The posters were presented on 3 foot by 6 foot boards in the lobby area between the two meeting rooms, the same area as the exhibits and break refreshments There were a couple of papers on SCIM in the poster session (including mine), and several papers on Parlay based systems. My paper proposed using an application server for SCIM, using a combination of rule based resolution and programmed components to mediate among different applications using different protocols. There was a Teleca paper on resolving service interactions among SCPs using IN using rules. I had an interesting discussion with the author and one of the founders of the company who said they had had good results in trials, but the first deployment wasn’t until August. The Parlay posters were from Universities and I never had a chance to talk with the authors.
There were
only two exibitors this year, Appium
and Teleca/Sonera.
I had an interesting discussion with the Appium
CTO. Appium
has been around for 10 years and has been involved in building applications for
various environments, mainly Parlay.
It’s not clear from what I heard how much was deployed in service and
how much trial, but they have had a lot of experience in dealing with feature
interactions and the limitations of various IN and service programming.
The Exhibits are one area where ICIN has shrunk dramatically since 1998, when my team was part of a Lucent booth exhibiting internet call waiting along with booths from Alcatel, Siemens, and several other large players. In part Exhibiting at ICIN seems less attractive than some other conferences because the focus is on the technical sessions and people are in the exhibits only during breaks. The setup this year though was quite nice with the exhibits prominent in the lobby, and in talking with others on the planning committee I learned that some of the smaller companies thought that ICIN was a very good venue for them because the major vendors don’t exhibit and as a result the smaller players can be a big part of the show. In 2003, Personeta had an exhibit staffed by Alon Mellor and me and I thought we had reasonable traffic.
This
conference has a substantial social program in addition to the technical
sessions. By now this has a bit of the
feel of an alumni reunion to me, knowing so many of the regular attendees and
the location and looking forward to finding out what everyone is up to. During those breaks and the receptions and
dinner it was clear that the telecom industry is still one that is under a lot
of stress. Alcatel sponsored a cocktail
reception, and the speaker for Alcatel (I was too late in arriving due to a
small problem with the busses to hear who she was) gave as introduction the
fact that Alcatel is seeing a decline in both its wireless and wireline businesses and is looking towards services and
service delivery solutions (the real topic of ICIN) as a new source of revenue
to replace that lost in other areas. The recent rounds of mergers (e.g. Alcatel/Lucent,
AT&T/SBC/BellSouth clearly have introduced uncertainty both into the
company plans for deployment and the career plans of their key employees. It is very unfortunate that at a time when
the industry most needs people focused on bold moves and new ideas so many
people are concerned more about their personal futures.
The City of
Bordeaux on the other hand is clearly enjoying an improving climate. Many buildings have been cleaned of centuries
of grime and the city continues to develop shopping and restaurants while
preserving its historic character. It’s
still very much a walking city, though a new tram system makes it possible to
get around easier than in the past. Well
worth a visit.
Kevin
presented a brief summary of the conference
Chairman’s
observations:
The Hot
topics for the conference:
Topics for
the next conference
Next ICIN
will be October 8-11 2007 in Bordeaux (Likely to be Cite Modial)
New TPC
chairman Ulrich Reber (Siemens) (Ulrich.Reber@siemens.com) and New
IAB chairman (Stuart Sharrock)