building scalable sip networks - ag projects · 5/18/2006 · ag projects building scalable sip...

AG Projects Building scalable SIP networks

May 18, 2006

Stockholm

Building scalable SIP networks

Adrian [email protected]


My name is Adrian Georgescu

• Founder and CEO of AG Projects

• Co-chair of Dutch SIP SIG working group

• Member of OpenSER management board

• Member of ETSI ENUM special task force


IETF standardization for the SIP protocol is nearly finished, the SIP protocoland related components are already defined

Current IETF activities around SIP:

1. Presence standardization will be finished in June 2007

2. ENUM - IP connectivity with E.164 numbers

3. SPEERMINT - peering policies between VoIP providers

4. P2P SIP - make SIP simple to use, solve scalability problems


So, what does the end-user want?

Connectivity via any device, anywhere, at anytime:

• global mobility

• personalization

• ease of use

• fair prices

• freedom to chose additional services


What do service providers want?

• Target end-users all over the world

• Small acquisition cost per end-user

• Minimum operational expenditure

• Keep pace with innovation and remain competitive

• Business scalability


How to deliver what users want?

Telecom or Internet approach?


PSTN is a centralized network where various elements are chained tocontrol access and perform billing

NGN proposed by ITU-T, 3GPP IMS and its ETSI TISPAN extensions aimto implement the same model using Internet protocols


• the network is application aware

• the control resides in the network

• NGN has many QoS definitions to “guarantee” for the network services

all ITU-T networks, such as ISDN, BISDN/ATM/NGN are based on granddesigns and not based on evolution.

The changes from TDM to ATM to IP are significant discontinuities in theITU-T architectures.

The ITU-T view on the network is:


Internet is a dumb network, the services are performed at the edgebased on the end-to-end principle

Internet reached current status through natural evolution


1. Network is unaware of the applications, this makes it scalable

2. Nothing should be done in the network that can be done in an end-system

3. Quality of Service (QoS) is a matter of bandwidth availability and not ofcentral control

... the end-to-end principle is the most fundamental Internet’s architecturalprinciples:


On the PSTN, voice services follow a “create and manage bottleneck”architecture, where arbitrary points of control are created to justifybusiness models based on resource scarcity


On Internet, services are available on the edge, can spread virally,business become successful only if end-users find the services orapplications interesting.

Internet is an “eat all you can” model based on flat fees


For real-time communication applications like VoIP, a client needs only:

1. Global routable identity (DNS, ENUM)

2. Address translation (SIP, NAT)


New business models are based on context aware communicationsreal-time communications linked with subscriptions/notifications toevents from personal and business spheres.

Presence framework developed by the IETF SIMPLE working groupenables the publication, subscription and notification to a broad rangeof information


Presence enables context aware communications:

the online/offline buddy status is the simplest example.

• enables mobility and tele-working

• integration between VoIP and World Wide Web

• ticket sale start for events like entertainment

• check stock items at your supplier

• communicate efficiently based on availability and environment


• 1 to N messaging model

• Reach end-points in parallel on different networks

• Frequent updates to multiple recipients

Presence and SIP parallel forking are not “channel based”. Both concepts fallout of the telephone switch design and all the NGN components that havebeen built around managing channels (like Session Border Controllers)


“Presence is the dial-tone of the 21-st century”

Henry Sinnreich, the godfather of SIP


1. New strategy, marketing and sales vision

2. Scalable solutions for SIP Proxy, Presence and NAT traversal

3. Possibility to combine hardware and software SIP applications

4. Understand what end-users want

To roll out new business models you need:


New business models target global Internet, are designed to scale tomillions of subscribers (like Google, Yahoo, MSN and of course Skype)

any technical bottleneck can hurt new business models


Some are building IMS (Internet Multimedia Subsystem)

What is IMS?


IMS architecture - a crash course reality tour

• Classic telephony design, enforcing resource scarcity

• End-to-end communication not possible between end-points

• At least 12 components with at least 22 interfaces, a forest ofblack-boxes that introduce delays, bottlenecks and interoperabilityproblems

• SIP technology artificially incompatible between vendors

• Upgrade of all components is needed to allow new applications inthe end-points, because IMS network is aware of applications

• Must be kept in a walled garden, software is not Internet proof

• Last but not least, difficult to scale


Try to build this:


3GPP IMS Architecture – all subsystems


Oth

er IP N

etwo

rks

IP Transport (Access and Core)

T-MGF

I-BGF

UPSF

P-CSCF

I/S-CSCF

BGCF

SLF

ChargingFunctions

IWF

PSTN Emulation (R2)

Mw

Mw/Mk/Mm

Mr

Mg

Mj

Mi

Mp Mn

GmGq'

ISCCx Dx

DhSh

Ic

Rf/Ro

Rf/Ro

Ib

Iw

Gq'

PS

TN

/ISD

N

SGFMRFC MGCF

MRFP

e4

Ie

MwIBCF

MkMk

Application ServersRf/Ro

AGCF

e2

P1

P2

P3

UE

CNG

MG

IMS /PSTN Simulation

Gq'

-

SPDF

A-RACF

Resource & AdmissionControl

Resource & AdmissionControl

SPDFNetwork

Attachment Subsystem

Re Ia

RCEF BGF

UtUt

ETSI TISPAN IMS Architecture


Session & Call Control

Application A - 2: A pplication Gateway F E

NNI

Other NGN

other IP MM Network

(e.g. IMS)

T - 6 : Traffic Measurement F E T - 3 : T.Network

Access Process FE

T - 13 : Access Relay FE

T - 1 6 : T. Authentication

&Authorization FE

T - 18 : T. User Profile FE

Transport T - 6 : Traffic

Measurement F E

T - 1 : Core Packet Transport Function s

T - 8 : Transport Resource &

Enforcement FE T - 9 :

Access Border

Gateway F E

PSTN/ISDN

T - 17 : T.Network Access Control FE

T - 21: I - TRCF

Scope of NGN

S - 5 : Media GW Control F E

Internet

gement functions

T - 12 : Edge Node FE

T - 5 : T runk Media

Gateway F E T - 1 1 : Access Packet Transport F unctions

T - 8: MBS - FE T - b: Multicast

M BS FE T - a: MM - FE

T - 8: MBS - FE T - b: Multicast

M BS FE T - a: MM - FE

T - 10 : Access Node FE

T - 4 : Access Media

Gateway F E

T - 2 : Packet

Gateway F E

T - 19: A - TRCF T - 20: C - TRCF

T - 8 : Transport Resource &

Enforcement FE

S - 3 : S. Authentication & Authorization F E S - 10 : Subscription

Locator FE

T - 15: PD FE

S - 9 : Breakout Gateway FE S - 8: Session Control

Proxy FE S - 7 : A ccess GW

Control F E

S - 1: Session Control FE

A - 1: Application Server F E (may include own Authentication, Authorization and Accounting)

S - 4 : Media Resource Control FE

UNI

Terminal Function s

IP address allocation Authentication Authorisation

Access net. config Location mgt.

NAAF IP address allocation

Authentication Authorisation


IP address allocation Authentication Authorisation


NAAF

T - 7 : Media

Resource Processing

FE

S - 2 : S. User Profile F E

Multimedia Service FE

T - 1 4 : S ignalling Gateway

F E

S - 6 : Packet GW Control F E

S - 11 : Interrogating Session Control FE

both based on ITU-T NGN System Architecture


More than seven years old, IMS is still not deployed, while its release trainlengthens with changes and expansions (IPTV is next). IMS complexityresults from:

• decomposing devices into most granular functions and links

• tracking and controlling user behavior

The proliferation of boxes and protocols for the state managementrequired for data tracking lead to cognitive overload but add littlevalue.


When ready, IMS will provide less services than people got from Skypeyesterday, but at a huge cost.

Are your customers willing to pay for this?


Many operators like Cable companies and Internet service providershave chosen a common-sense and cost effective solution, the basicIETF model:


• Business model

• Secure identity

• NAT traversal

• How to scale

IETF model is simple to understand and build. You only need to address:


• Basic VoIP functionality is free

• Pay for the SIP device and get a free software for computers

• Pay for identity (Domain name or SIM card)

• Pay for value-added services

Business model?


How to solve the scalability problem?

How to avoid scaling up your cost of operations while connectingmore and more end-users?


Today we have enough computing power to run on a single server:

1. SIP Proxy and Registrar

2. Media relay

3. DNS and ENUM

4. Presence and XCAP

5. User profile

A single server can handle 10K users and 200 media sessions!


Take only the IMS functions that you really need


Fold all functions into a single box:


Play the hat-trick:

Scale the ”IMS in a Box“ using a self-organizing overlaynetwork by using Peer-to-Peer technology


Thank you,

Adrian Georgescu

[email protected]

http://ag-projects.com

building scalable sip networks - ag projects · 5/18/2006 · ag projects building scalable sip...

Documents