Upload
david-webster
View
213
Download
0
Tags:
Embed Size (px)
Citation preview
“Improving the scalability of an IMS-compliant conferencing framework through presence and event notification”
IPTComm2007New York, July 19th 2007
A. Amirante, T. Castaldi, L. Miniero and S.P. Romano
COMICS (COMputers for Interaction and CommunicationS)
research groupUniversity of Napoli Federico II
IPTComm2007 -- New York City, July 19th 2007
2
Agenda
Rationale & Motivation A standard centralized conferencing
architecture Performance assessment and bottlenecks
identification Improving scalability through distribution of
components/responsibilities Conclusions
IPTComm2007 -- New York City, July 19th 2007
3
Conference
The term “Conference” can be used to describe any meeting of people that “confer” about a certain topic Web Conferencing is used to indicate live meetings or presentations over the Internet
IPTComm2007 -- New York City, July 19th 2007
4
Standardization efforts
No standardization for many years Lack of interoperability Platform dependency Security issues Cost Market segmentation
Standardization Bodies ITU (International Telecommunication Union) IETF (Internet Engineering Task Force) 3GPP (3rd Generation Partnership Project)
IPTComm2007 -- New York City, July 19th 2007
5
IETF SIPPING Working Group Session Initiation Proposal Investigation Documents the use of SIP for several applications
related to telephony and multimedia SIP Conferencing
Loosely-Coupled Conference Fully Distributed Multiparty Conference
Tightly-Coupled Conference
Focus Policy Server Mixer
Notification Service (Event Package, RFC 4575) Participants
SIP Conferencing Framework (RFC 4353): fundamental elements
IPTComm2007 -- New York City, July 19th 2007
6
XCON Working Group
Centralized Conferencing (XCON) Extends RFC 4353
Protocol-agnostic (not only SIP) Data Sharing (not only audio/video)
Suite of Protocols Conference Control (CCMP?) Floor Control (BFCP) Call Signaling (SIP/H.323/IAX/etc.) Notification (Event Package?)
IPTComm2007 -- New York City, July 19th 2007
7
XCON Framework
Conference Control Client
Floor Control Client
Call Signaling Client
Notification Client
Conference Control Server
FloorControl Server Focus Notification
Service
Conference Object
Conference Object
Conference Object
Conference Control Protocol
Floor Control Protocol
CallSignaling Protocol
NotificationProtocol
Conferencing System
Conferencing Client
IPTComm2007 -- New York City, July 19th 2007
8
CONFIANCE CONFerencing IMS-enabled Architecture for
Next-generation Communication Experience http://confiance.sourceforge.net/
Open source implementation of the XCON Framework, compliant with the IMS specification
Extends the Asterisk PBX functionality Enhanced “MeetMe” application
Support for Conference Management (Scheduler) Support for Floor Control (BFCP) Support for BFCP-guided video switching/mixing Support for MSRP (Message Session Relay Protocol)
text chatrooms
IPTComm2007 -- New York City, July 19th 2007
10
Why Distributed Conferencing? Centralized Conferencing
Poorly scalable Limited capabilities Single point of failure
Towards distribution: Cascaded Conferencing
Each focus is seen as a participant by the others Only affects mixers' distribution Centralized protocols like BFCP don't work
IPTComm2007 -- New York City, July 19th 2007
11
DCON Proposal Distributed Conferencing (DCON)
http://dcon.sourceforge.net/ Orchestrates the operation of a set of XCON “clouds” Overlay network interconnecting the clouds Intra-focus communication
Still based on XCON protocols Inter-focus communication
Exploits Server-to-Server (XMPP) Requirements
Focus discovery Initialization information & spreading of conference events Setup and managing of distributed conferences Transparent dispatching of natively centralized protocols
among the involved conferencing clouds
Focus DCON
Minisip Spark
Integrated client
XMPP
SIP
IPTComm2007 -- New York City, July 19th 2007
14
Tool employed: SIPp SIP traffic generator Capable to reproduce fully customizable call scenarios through
XML configuration scripts Capable to also reproduce actual media flows through the
RTP/pcap replay function
Used to test two different facets of scalability: Maximum number of users who can access the system Resource consumption, given a specific number of users
Testbed:OS: Linux Fedora Core 6, kernel 2.6.15CPU 3,2GHz Intel XEON, RAM 2GB
Performance assessment
IPTComm2007 -- New York City, July 19th 2007
15
Asterisk does not allow to open more than 1024 Zaptel pseudo-channels (i.e. abstractions of physical channels) simultaneously
Each conference, upon creation, makes use of two such channels
Each conference participant uses a single pseudo-channel
Quasi-linear growth in the number of users in the presence of distribution
Note well: No RTP traffic involved in such tests just signaling!
Scalability: just signaling
IPTComm2007 -- New York City, July 19th 2007
16
0 50 100 150 200 250 3000
10
20
30
40
50
60
70
80
90
100
Calls
CP
U lo
ad [
%]
Centralized case (one single conference)
Focus Calls CPU load (%)
Main 300 99,4
300 users as a maximum limit in the presence of audio streams
Might become critical in case many conferences are activated in a cloud
A reference value for the following tests
Scalability: signalling + audio
IPTComm2007 -- New York City, July 19th 2007
17
0 50 100 1500
10
20
30
40
50
60
70
80
90
100
Local calls
CP
U lo
ad [
%]
Main focusRemote focus
Focus Calls CPU load (%)
Main 150 30,04
Remote 150 20,19
Scalability: 2 islands case
IPTComm2007 -- New York City, July 19th 2007
18
0 50 100 1500
10
20
30
40
50
60
70
80
90
100
Local calls
CP
U lo
ad [
%]
Main focusRemote focus 1Remote focus 2
0 10 20 30 40 50 60 70 80 90 1000
10
20
30
40
50
60
70
80
90
100
Local calls
CP
U lo
ad [
%]
Main focusRemote focus 1Remote focus 2
Focus Calls CPU load (%)
Main 100 20
Remote_1 100 18
Remote_2 100 18
Focus Calls CPU load (%)
Main 150 31,05
Remote_1 75 12
Remote_2 75 12
Scalability: 3 islands case
IPTComm2007 -- New York City, July 19th 2007
19
Migration towards a distributed paradigm allows for a huge reduction in the load of the primary focus
The sum of the CPU levels of all involved foci is less than the CPU level of the single focus in the centralized case
Given a fixed number of local users, remote users distribution among multiple islands adds negligible overhead to the main focus
Number of islands
Number of local users
Number of remote users
Main focus CPU load
Remote focus 1 CPU load
Remote focus 2 CPU load
Remote focus 3 CPU load
1 300 - 99.4% - - -2 150 150 30.04% 20.19% - -3 100 200 (100/100) 20% 18% 18% -3 150 150 (75/75) 31.05% 12% 12% -4 75 225 (75/75/75) 12.66% 12% 12% 12%4 150 150 (50/50/50) 32.4% 7.8% 7.8% 7.8%
Scalability: figures in summary
IPTComm2007 -- New York City, July 19th 2007
20
Future Work
More performance tests: Different media:
Involve video, chat, desktop sharing applications, etc. in the trials
Different protocols Involve BFCP, Conference Control Protocol, etc. in the
trials Different scenarios
Try to reproduce real-world conference configurations Different overlay topologies
Not just full meshes Investigate smart p2p approaches
IPTComm2007 -- New York City, July 19th 2007
21
What else?
See all of you at the demo tomorrow! Have a look at DCON up and (hopefully ) running… We also present:
our brand new JAVA-based integrated client our current prototype of the distributed approach to video
mixing
Our contacts: [email protected] [email protected] [email protected] [email protected]