Audio networking Franois Dchelle (dechelle@ircam.fr) Patrice - - PowerPoint PPT Presentation

audio networking
SMART_READER_LITE
LIVE PREVIEW

Audio networking Franois Dchelle (dechelle@ircam.fr) Patrice - - PowerPoint PPT Presentation

Audio networking Franois Dchelle (dechelle@ircam.fr) Patrice Tisserand (tisserand@ircam.fr) Simon Schampijer (schampij@ircam.fr) IRCAM LAD 2004 Karlsruhe / 1 Distributed virtual concert project and issues network protocols and


slide-1
SLIDE 1

LAD 2004 – Karlsruhe / 1

François Déchelle (dechelle@ircam.fr) Patrice Tisserand (tisserand@ircam.fr) Simon Schampijer (schampij@ircam.fr) IRCAM

Audio networking

slide-2
SLIDE 2

LAD 2004 – Karlsruhe / 2

Distributed virtual concert

project and issues network protocols and libraries nJam

Experimentations in grid computing

what it is middlewares for grids tests and benchmarks

slide-3
SLIDE 3

LAD 2004 – Karlsruhe / 3

Distributed virtual concert

Goal: to distribute a concert (musicians, sound engineers and public) on the Internet Performance is geographically distributed Real time audio streaming over Internet Developed technologies can be applied to other domains (virtual or augmented reality distributed applications) A joint project between Ircam and CNAM

slide-4
SLIDE 4

LAD 2004 – Karlsruhe / 4

Similar projects

SoundWIRE (CCRMA): using networks with high QoS for musical collaborations GRAME: distributed rehearsal, network protocol for events streaming, clock skew compensation Many others, using ISDN or MIDI over internet

  • r OGG or ...
slide-5
SLIDE 5

LAD 2004 – Karlsruhe / 5

Issues

Network adds a latency that varies a lot Latency is difficult to manage for a musician Real time and transfer reliability are difficult to

  • btain together

n sources and m destinations means high traffic and requires native multicast Clock skew between sources and destinations

slide-6
SLIDE 6

LAD 2004 – Karlsruhe / 6

Network protocols for audio

RTP/RTCP

real-time protocol for audio and video applications RTP uses UDP as underlying layer provides packets typing and timestamping RTCP provides control and reporting

Other protocols of interest:

SIP/SDP: Session Initiation/Description Protocol RTSP: Real Time Streaming Protocol

slide-7
SLIDE 7

LAD 2004 – Karlsruhe / 7

Libraries and applications

Some libraries that implement RTP:

liveMedia (www.live.com/liveMedia) ccRTP (www.gnu.org/software/ccrtp/ccrtp.html) Vovida (www.vovida.org) UCL Common (www-mice.cs.ucl.ac.uk/multimedia/software)

Some applications that support RTP:

VideoLan Mplayer

slide-8
SLIDE 8

LAD 2004 – Karlsruhe / 8

nJam

An implementation of the distributed virtual concert inside jMax with the UCL library Multicast support Each participant sends local sound on a multicast address and receives all participants streams A distributed algorithm compensates network latency and resynchronizes locally all streams Musicians hear themselves with a fixed delay

slide-9
SLIDE 9

LAD 2004 – Karlsruhe / 9

nJam experimentation

Experimentation of a distributed concert during Résonances 2003 at Ircam 4 musicians on 2 sites Soft multicast using a tunnel Musicians trained fast during rehearsal Public presentation failed:

port scanning attack just before concert no QoS network giving priority to audio packets

slide-10
SLIDE 10

LAD 2004 – Karlsruhe / 10

Jack RTP client

Goals:

implement distributed virtual concert outside jMax provide a networking Jack client using standard protocols

First development using LiveMedia library Multicast support Under development...

slide-11
SLIDE 11

LAD 2004 – Karlsruhe / 11

Distributed virtual concert: future

Finish the reimplementation using Jack Add control to the Jack client (with which protocol?) Implement clock skew compensation (difficult issue: breaks the synchronization algorithm) Make large scale tests Consider xDSL (compression, no multicast)

slide-12
SLIDE 12

LAD 2004 – Karlsruhe / 12

Grid computing

Virtualization of computing and data resources

  • ver high speed networks

Unlike clusters, grids can be geographically distributed and heterogeneous Used mainly in high performance computing Based on open standards and protocols, the Open Grid Services Architecture (OGSA)

slide-13
SLIDE 13

LAD 2004 – Karlsruhe / 13

Grid computing for audio

Audio is less demanding than particle physics, but grid computing can help to ... get more CPU than your single desktop ... process large sound databases ... use applications remotely ... explore large parameters spaces

slide-14
SLIDE 14

LAD 2004 – Karlsruhe / 14

Ircam grid testbed

Small deployment for tests and benchmarks « cycle stealing » on standard workstations Linux only for now, Mac OS X soon 2 middlewares tested:

Globus toolkit (http://www.globus.org/) XtremWeb (http://www.xtremweb.org/)

Future deployment : ~ 60 machines

slide-15
SLIDE 15

LAD 2004 – Karlsruhe / 15

The Globus toolkit

A free software toolkit for building grids The base of many grid deployments Features:

grid security remote job submission and control high-performance secure data transfer system and service information

Heavy, but efficient, standard and portable

slide-16
SLIDE 16

LAD 2004 – Karlsruhe / 16

XtremWeb

A free software platform for global computing Architecture:

servers dispatch jobs workers execute jobs dispatched by servers clients submit and control jobs

Much lighter than Globus, portable, but less efficient than Globus Still under development

slide-17
SLIDE 17

LAD 2004 – Karlsruhe / 17

Globus vs. XtremWeb

Example using Csound: On a 4 machines grid, speedup is about 2 Disappointing, but can be explained...

slide-18
SLIDE 18

LAD 2004 – Karlsruhe / 18

Issues

Globus is too heavy for a large deployment on a grid of desktop machines XtremWeb is not enough stable and efficient Our testbed is too small Heterogeneity is difficult to manage But it offers interesting possibilities

slide-19
SLIDE 19

LAD 2004 – Karlsruhe / 19

Grid computing: future

Distributed computation by exploding each computation into a set of jobs communicating via MPI:

MPI speed can be an issue not all applications can be parallelized this way

Merging of grid services and web services (Globus toolkit 4) Integration with streaming architecture