Multicast Multimedia Traffic and the Design of Real-Time Protocols Bruce A. Mah bmah@tenet.Berkeley.EDU The Tenet Group University of California at Berkeley and International Computer Science Institute Y • O T I F S • R C A E V A L I I N F O U • R L E E L I G T H T N H T H I E R B E A T E • • • 1 8 • 8 6 16th Annual ILP Conference 10 March 1994 Multicast Multimedia Traffic and the Design of Real-Time Protocols Page 1 of 16
Outline Introduction and motivation Environment and methodology Measurements Analysis and other observations Conclusions and implications for protocol design Multicast Multimedia Traffi c and the Design of Real-T ime Protocols Page 2 of 16
Introduction IP Multicast MBONE (Multicast Backbone): Virtual network for supporting Internet-wide multicasts Multicast “sessions” and session directory Multimedia tools Video conferencing ( nv, vic, ivs ) Audio conferencing ( vat, nevot ) Shared whiteboard ( wb ) Still image distribution ( imm ) Questions What can we learn about network traffi c generated by production use multimedia applications? How can real-time network protocols best support these applications? Multicast Multimedia Traffi c and the Design of Real-T ime Protocols Page 3 of 16
Environment and Methodology tcpdump To XUNET MBONE XCS (Ethernet) routers XUNET 128.32.131 192.128.57 propaganda paradigm 192.107.102 conviction Tenet (FDDI) To campus 128.32.201 and Sequoia 2000 ICSI (Ethernet) MBONE routers Capture packet headers for off-line processing sdsnoop : Session Directory Snoop Multicast Multimedia Traffi c and the Design of Real-T ime Protocols Page 4 of 16
Aggregate IP Multicast Traffi c XUNET Video Conference 22 September 1993 11:34 AM to 1:06 PM PST 2000 Total IP Multicast Traffic 1500 KBits per second 1000 500 0 0 1000 2000 3000 4000 5000 6000 Time in Seconds Multicast Multimedia Traffi c and the Design of Real-T ime Protocols Page 5 of 16
Aggregate IP Multicast Traffi c ”Just an ordinary day” 20 January 1994 09:45 AM to 21 January 1994 09:47 AM PST 2000 Total IP Multicast Traffic 1500 KBits per second 1000 500 0 0 10000 20000 30000 40000 50000 60000 70000 80000 90000 Time in Seconds Multicast Multimedia Traffi c and the Design of Real-T ime Protocols Page 6 of 16
Audio Bitrate ( vat ) XUNET Audio 22 September 1993 11:34 AM to 1:06 PM PST 140 120 KBits per second 100 80 60 40 20 0 0 1000 2000 3000 4000 5000 6000 Time in Seconds Multicast Multimedia Traffi c and the Design of Real-T ime Protocols Page 7 of 16
Video Bitrate ( nv 128Kbps) law.CS.Berkeley.EDU to XUNET Video (one-second samples) 22 September 1993 11:34 AM to 1:06 PM PST 300 law.cs.berkeley.edu 250 KBits per second 200 150 100 50 0 0 500 1000 1500 2000 2500 3000 3500 4000 Time in Seconds Multicast Multimedia Traffi c and the Design of Real-T ime Protocols Page 8 of 16
Video ( nv ) Bitrate law.CS.Berkeley.EDU to XUNET Video (fi ve second samples) 22 September 1993 11:34 AM to 1:06 PM PST 300 law.cs.berkeley.edu 250 KBits per second 200 150 100 50 0 0 500 1000 1500 2000 2500 3000 3500 4000 Time in Seconds Multicast Multimedia Traffi c and the Design of Real-T ime Protocols Page 9 of 16
Video ( nv ) Packet Sizes law.CS.Berkeley.EDU to XUNET Video 22 September 1993 11:34 AM to 1:06 PM PST 1 Cumulative Distribution 0.8 0.6 0.4 0.2 0 0 200 400 600 800 1000 1200 1400 Packet Size in Bytes Multicast Multimedia Traffi c and the Design of Real-T ime Protocols Page 10 of 16
Video ( nv ) Packet Interarrival Time law.CS.Berkeley.EDU to XUNET Video 22 September 1993 11:34 AM to 1:06 PM PST 1 Cumulative Distribution 0.8 0.6 0.4 0.2 0 0 20 40 60 80 100 120 140 160 180 200 Interarrival Time in Milliseconds Multicast Multimedia Traffi c and the Design of Real-T ime Protocols Page 11 of 16
Still Image Distribution ( imm ) GOES-7 (Visual Satellite Images) 10 8 KBits per second 6 4 2 0 0 10000 20000 30000 40000 50000 60000 70000 80000 90000 Time in Seconds The solution to “casual Internet usage?” Multicast Multimedia Traffi c and the Design of Real-T ime Protocols Page 12 of 16
Some Traffi c Analysis 20 January 1994 Trace Gross Characteristics 299 traffi c sour ces 65 destinations 1,005 conversations 722,901,051 total bytes 2,239,382 total packets No special events this day...why so many destination addresses? Multicast Multimedia Traffi c and the Design of Real-T ime Protocols Page 13 of 16
Traffi c Br eakdown 723 MB total (all sessions) 585 MB (81.0% percent of total) from a locally-advertised “radio session” , adjusted totals exclude this session 24 advertised sessions 100 MB (72.6% of adjusted total) Still Images ( imm ): 46 MB Audio ( vat ): 42 MB Video ( nv ): 11 MB 7 unadvertised but known sessions 8 MB (5.5% of adjusted total) 33 unknown sessions 30 MB (21.8% of adjusted total) Multicast Multimedia Traffi c and the Design of Real-T ime Protocols Page 14 of 16
Funny User Behavior What’s happening here? c patterns suggest vat audio conferencing Traffi Lack of session directory advertisements suggests testing, experimentation Improper scope control (most participants in Europe, why should we see their traffi c in California?) Suggestions We need real multicast tree pruning! Users need education! Hosts unreachable from NSFNET backbone Multicast Multimedia Traffi c and the Design of Real-T ime Protocols Page 15 of 16
Conclusions and Implications Aggregate traffi c IP Multicast traffi c still highly variable Dependent on special events and outside factors Diffi cult to constr uct a “typical” workload Flexibility is important Audio conferencing traffi c “Floor control” needed c ( nv coded) Video conferencing traffi Peak-to-average ratio of about 2:1 Large packets sent fairly frequently User behavior Better protection in network needed (true multicast tree pruning) User education Multicast Multimedia Traffi c and the Design of Real-T ime Protocols Page 16 of 16
Recommend
More recommend
