MMCN’05 Exploiting Content-Based Networking for Fine Granularity Multi-Receiver Video Streaming Viktor S. Wold Eidea,b Frank Eliassena Jørgen Andreas Michaelsenb a Simula Research Laboratory , Lysaker, Norway {viktore,frank}@simula.no b University of Oslo , Norway {viktore,jorgenam}@ifi.uio.no http://www.ifi.uio.no/˜dmj/ http://www.simula.no:8888/QuA/ Twelfth Annual Multimedia Computing and Networking (MMCN ’05) January 19-20, 2005, San Jose, California, USA Part of Electronic Imaging Symposium Simula Research Laboratory and University of Oslo, Norway 1 of 19 MMCN’05 Motivated by observing that multi-receiver video streaming still represents a challenge Unicast solutions, only big players can provide live video to many receivers Network layer multicast is still not ubiquitous However, if multicast was ubiquitous, heterogeneity would still be a challenge Simula Research Laboratory and University of Oslo, Norway 2 of 19 MMCN’05 The goal is to allow each video receiver to customize the video quality independently A video receiver may then balance resource usage against cost and preferences The number of interested video receivers should not affect the video server The network should be relatively unaffected by the number of video receivers Simula Research Laboratory and University of Oslo, Norway 3 of 19 MMCN’05 Our suggestion is to break the traditional assumption of destination (group) addressing Traditionally, a video server adds an explicit destination (group) address to each outgoing packet. The address then determines the forwarding and delivery In contrast, our approach is to use implicit addressing, where the contents of each packet and the receiver interests determine the delivery Hence, we suggest exploiting content-based networking for fine granularity multi-receiver video streaming Simula Research Laboratory and University of Oslo, Norway 4 of 19 MMCN’05 The rest of the talk presents the bridging of content-based networking and video streaming The following will be discussed: • Content-based networking • An encoding scheme for fine granularity video streaming • Video streaming demonstration • Empirical results • Conclusion and further work Simula Research Laboratory and University of Oslo, Norway 5 of 19 MMCN’05 Content-based networking have some distinguishing characteristics Messages, called notifications, are forwarded based on content, and not on an explicit address. The notifications: • consist of attribute/value-pairs: [ sid=10 col=1 row=2 blob=... ] • are delivered to interested clients based on subscriptions: [ sid=10 col>=2 ] • are pruned upstream and replicated downstream • may be transported over different technologies, such as TCP, UDP, and IP multicast Simula Research Laboratory and University of Oslo, Norway 6 of 19 MMCN’05 Content-based networking example, including a single video server and some video receivers Content−Based Network Publish Video server Video client(s) Network node Intra domain Subscribe Notify Subscribe Subscribe Notify Notify Simula Research Laboratory and University of Oslo, Norway 7 of 19 MMCN’05 A fine granularity coding scheme allows each video receiver to customize the video stream Well known video coding techniques have been adopted to provide each video receiver with independent selectivity along multiple video quality dimensions: • region of interest (superblocks) • signal to noise ratio (DCT coefficients quantized and viewed as bit-planes) • colors (separating luminance and chrominance information) • temporal resolution (layered structure) Simula Research Laboratory and University of Oslo, Norway 8 of 19 MMCN’05 Region of interest selectivity is supported by means of so-called superblocks Region of interest selectivity row col 0 1 0 1 n superblock macroblock block Y0 Y1 Y2 Y3 U V m Simula Research Laboratory and University of Oslo, Norway 9 of 19 MMCN’05 SNR selectivity is supported by means of bit-plane coding Signal to noise ratio selectivity bit ql sign 10 MSB ... MSB − 1 0 0 Advertise 1 2 3 0 1 2 62 63 ... DCT values of 8x8 block Simula Research Laboratory and University of Oslo, Norway 2 1 0 LSB 10 of 19 MMCN’05 Each frame is broken up, encoded, and published in a number of notifications Video fragments ... Publish Notifications sid=1 tl=2 ql=0 f=0 row=2 col=3 blob=... sid=1 tl=2 ql=3 f=1 row=1 col=3 blob=... Video server Simula Research Laboratory and University of Oslo, Norway 11 of 19 MMCN’05 The content-based network forwards the notifications based on subscriptions Notify CB Subscriptions Map Publish CB IP Subscriptions Map IP Notify CB 234.0.80.3 234.0.80.2 Subscriptions Map IP Network Simula Research Laboratory and University of Oslo, Norway 12 of 19 MMCN’05 Each notification is delivered to all receivers having subscriptions matching the notification Notify First Video Client Notify Second Video Client Video Receivers Simula Research Laboratory and University of Oslo, Norway 13 of 19 MMCN’05 Video streaming demonstration Simula Research Laboratory and University of Oslo, Norway 14 of 19 MMCN’05 A layer of indirection is provided between the video server and the video receivers CB Notify Subscriptions Map CB IP Subscriptions ... Publish Map CB Video fragments IP sid=1 tl=2 ql=0 f=0 row=0 col=3 data=[...] Notifications 234.0.80.3 234.0.80.2 Notify Subscriptions Map sid=1 tl=2 ql=3 f=1 row=1 col=1 data=[...] IP Video server First Video Client Network Simula Research Laboratory and University of Oslo, Norway Second Video Client 15 of 19 MMCN’05 Experimental results indicate that network and CPU usage scales well By reducing the video quality in one or more dimensions, the resource consumption is reduced accordingly Hence, a video receiver may independently tradeoff resource usage against video quality in the different dimensions Simula Research Laboratory and University of Oslo, Norway 16 of 19 MMCN’05 When receiving only some regions, the CPU and network consumption drops accordingly† News CIF Y, 30 frames pr. second col row % CPU bpp intra bpp diff kbps 0 0 9.50 0.70 0.16 116 0 1 11.82 1.20 0.29 208 0 2 5.79 1.52 0.03 114 1 0 7.08 0.88 0.20 147 1 1 11.99 1.34 0.40 268 1 2 8.17 1.44 0.12 149 Sum 54.35 1.18 0.20 1002 All 55.37 1.18 0.20 1003 † Results for the other dimensions are in the paper Simula Research Laboratory and University of Oslo, Norway 17 of 19 MMCN’05 Content-based networking seems promising for fine grained multi-receiver video streaming The contribution of this work is bridging content-based networking with techniques from the fields of video compression and streaming Each video receiver may independently customize region of interest, colors, SNR, and temporal resolution, while efficiency (network and CPU) is maintained The software, written in Java, is available as open source from the project web pages Simula Research Laboratory and University of Oslo, Norway 18 of 19 MMCN’05 There are many opportunities and challenges with respect to further work We are currently working on: • motion compensation • adaptation, e.g., by automatically changing subscriptions • spatial scalability We are also considering: • closer integration with MPEG-4 • storage formats suited for non real-time usage, e.g., video on demand Simula Research Laboratory and University of Oslo, Norway 19 of 19