By Hui Ma, China Internet Weekly
August 20, 2009 2:00 PM PT
August 2008 was obviously the most hectic month of an exciting and proud summer for Beijing, China. But Yusuo Hu and the colleagues on his team were soaked with a hair-raising sense of tension all through that month. “Everybody was under immense pressure because of the live video broadcast service that was being offered on such a large scale for the first time ever. During the two weeks before and after the Olympic grand opening, team members worked in shifts at the CCTV.com office building until midnight,” said Hu, Associate Researcher of Media Communication Group at Microsoft Research Asia and the owner of the MSRA P2P Broadcast System Project, still vividly remembering the days he spent working on the joint project between Microsoft Research Asia and CCTV.com.
Development of the P2P Broadcast System got underway in September 2007 at Microsoft Research Asia, and was subject to a massive trial during the CCTV Spring Festival Evening and the European Cup in 2008. The project was later deployed at CCTV.com (official website of CCTV International) to feed live broadcasting content to more than 2 million netizens during the Beijing Olympics. From Microsoft Research Asia to the CCTV.com-MSRA Collaboration on Digital New Media, MSRA P2P Broadcast System is getting closer to the lives of ordinary people. For Hu, the summer of 2008 was only a “successful start” of P2P streaming exploration and innovation.
The value of basic research at Microsoft goes far beyond the bottom line; this is no different in the field of video.
Video content is streamed across the Internet in mainly two ways, CDN (Content Delivery Network) that feeds content to users from streaming servers, and P2P (Peer to Peer) that delivers video through self-organized P2P networks.
“Unlike CDN, P2P is technically capable of serving any number of users while saving a remarkable amount of bandwidth, which brings significant advantages to either content providers or TV stations, as it cost them much less to broadcast content over the Internet. They simply no longer need that much bandwidth and that many servers.”
Technologies such as PPLive, PPStream, and QQLive have been doing very well in the P2P streaming video broadcast market, but Microsoft still has its own views on unleashing the maximum potential of this technology.
“China has 338 million Internet users, but the network infrastructure here is quite different from those used in the United States and Europe, so P2P streaming technology will enjoy a rosy future here,” said Hu. According to him, Microsoft Research Asia has kept an eye on P2P streaming media since it first began receiving public attention in 2005, and the collaboration it had with CCTV.com was merely an open tryout.
The live broadcast of the CCTV Spring Festival Evening 2008 was a large-scale, pre-Olympic exercise. Hu and his team took the trial very seriously as it was the first time that this laboratory creation would meet the real world. Despite their countless debugging and testing efforts, Hu and his colleagues still frequently spotted frozen frames during the live broadcast. The trouble, detected at an outbound port in a certain server rack, turned out to be a mismatch between the planned bandwidth of the servers and the speed of routers they were linked to. The problem was solved when they reduced the number of server connections, but Hu learned his lesson from the incident. “I really underestimated the pitfalls between a research prototype and a mature application,” he said.
Three important indicators are used to evaluate the performance of a P2P streaming system: 1) Bandwidth utilization, or the highest bit rate possible within a given bandwidth distribution; 2) Latency, including the transmission delay (time lag between video source and client-end) and start-up delay (amount of time elapsed between user login and start of transmission); 3) Continuity.
Correspondingly, Microsoft Research Asia made three improvements, the first being distributed NAT traversal technology that enhances bandwidth utilization. “If household gateways are applied, users may hide behind NAT or firewalls, which makes it difficult for interconnection among them,” Hu explained, adding that some users on public networks (not behind firewalls) may serve as intermediate nodes that help those behind firewalls establish connections. Bandwidth behind the NAT/firewall is thus made accessible.
To possibly reduce delays, Microsoft Research Asia introduced contribution-aware mechanism for P2P networks. Given the varying capability of users’ computers, contribution-aware mechanism, through carefully designed algorithms, relocates nodes with relatively large bandwidths closer to video sources and nodes with relatively smaller bandwidths to edges of the P2P network. “A machine with relatively large bandwidth is able to quickly forward the data it receives to more downstream nodes, which shortens the radius of the P2P network, greatly reduces delay and improves bandwidth utilization. When a user watches video content, the P2P packet-scheduling algorithm determines the origins and volume of data to be requested from other users according to the users’ bandwidth condition so as to ensure smooth video playback. In the deployment-driven experiments, this optimized scheduling algorithm enables more than 95 percent of users to play video smoothly.
Everything Hu is doing at Microsoft Research Asia is more or less related to colorful images and videos. In fact, he has been engaged in studies concerning multimedia technology since the days of his internship.
Hu was profoundly intoxicated by his first assignment as an intern at Microsoft Research Asia -- something that went by the name of “Attention Model.” During the transmission of a very large image, the recipient usually sees it appear one small portion at a time instead of all at once. So Hu’s mentor asked him to figure out an algorithm that gave priority to the most interesting portions of an image – a person’s face, for example. When revisited, projects carried out in his early days at Microsoft Research Asia look so easy to Hu, but they did teach him his first lesson on the importance of user experience.
“Today’s P2P applications are still not that convenient as they require downloading and the installation of certain plug-ins. Standard components such as those designed for browsers or Flash have yet to be developed, but user experience is bound to be improved alongside the progressive popularization and standardization of P2P technology -- plug-ins will no longer be necessary.”
There is still a long way to go for P2P streaming technology. A primary reason is that current P2P applications usually bring heavy traffic to Internet Service Providers (ISPs). To broaden the future adoption of P2P streaming technology, Hu’s team is now working on reducing the cross-ISP traffics of P2P applications by some Internet measurement technology. Moreover, P2P-related development may involve a number of copyright issues, but these may be solved one at a time through improvements in the business model.
“Currently, a major bottleneck in P2P live video broadcasting is in image quality – it can’t be compared with that of HDTV. This can be partially blamed on constraints in network bandwidth, but we are already considering the feasibility of enabling better image quality through a hybrid P2P/server solution,” said Hu, who, along with his colleagues, won’t stop working until total user satisfaction is achieved.