Lessons Learned from the Live Dance Performance Synchronization Between Tokyo, New York, and London Over R&E Networks
Time 10/17/18 11:00AM-11:10AM
We will discuss the technical challenges required for the sophisticated real-time remote collaborative arts over high-speed network based on the trial we’ve done connecting Tokyo, New York and London last November. The application of the trial was a remote collaborative dance performance spanning the three cities in collaboration with R&E networks, Internet2, JISC and GEANT among others.
The video images of live performances of each dancer in New York and London were transmitted to Tokyo site, then the performance images shot in three locations including Tokyo were precisely synchronized to create a virtual unit dancing image using the low-latency Advanced MMT (MPEG Media Transport) transmission unit developed by NTT Laboratories.
The following results were obtained from the trial;
1. For this specific trial, the target latencies at the application level (from video camera to display) was set at 1 second so that the dancers performing at distant locations could perform together with relatively small difficulties.
2. At the IP network layer, the one way delay between Tokyo and New York was 80 milliseconds and the one between London and Tokyo was 110 milliseconds. The throughputs between those cities were nearly 1Gbps over the 1Gbps L2 path created over Internet2 (AL2S), GEANT (GEANT Plus), JISC (JANET) and GEMnet2 (NTT). The maximum end-to-end jitters were less than 1 millisecond for the both paths. At both ends of the links we used commercial subscriber services (Metro Ethernet) and the detailed performance monitoring was not conducted.
3. A few packet losses were observed during the long term traffic tests. We classified the patterns of the losses and tried to solve the issue. It turned out that the random packet loss (a few losses every few minutes) occurred at a specific switch model when applied the traffic from the MMT unit. Other loss patterns with more bursty nature could not be tracked down in the duration of long term test due to the lack of time.
4. Because we could not completely solve the issue before the actual trial, we applied a heavy FEC (30% of repair packets) on video / audio contents using the advanced MMT to prevent any losses at the application layer.
5. The synchronization of the video / audio signals from three sites was made possible with the precise time source at each location synchronized to GPS and CDMA signals. The final video / audio communications channel created over the network had the latency of 860 milliseconds with a complete frame synchronization (lag with less than 1 frame). This value could have been lowered with more monitoring and fine tuning of the parameters of each layer (network/middleware/application).
6. It has been proved that the sophisticated remote collaborative performance over very long distances can be achieved with high performance communications network when used in conjunction with advanced video / audio transport system (Advanced MMT in this trial) and precise time synchronization mechanisms.
7. We plan to continue our research and development to be able to include more complex collaborative art involving more number of people and locations. Especially the application delay issues caused by network performance indices will be pursued by controlling the packet loss and jitter performances to determine the optimum error correction strength.
A short video clip of the final synthesized result will be presented if the allocated time allowed.
Speaker Hisao Uose NTT (Nippon Telegraph and Telephone Corporation)
Speaker Tomoyuki Hayasaka NTT (Nippon Telegraph and Telephone Corporation)
Primary track Advanced Networking
Secondary tracks Advanced Networking