Christian Spielvogel

Collaborative Quality-of-Service based video conferencing

Video conferencing typically involves multiple parties who communicate over the Internet by exchanging real-time video and audio messages. The problem about Internet based real-time communication is the best effort behavior which sometimes causes erroneous transmissions that frustrate the conferencing parties. To avoid erroneous transmissions, Quality-of-Service aware video conferencing systems dynamically adapt to varying network conditions for the price of decreased video frame rate, resolution or quality. In this paper we go one step beyond Quality-of-Service awareness and present a collaborative approach for achieving the required quality in video communications that does not require adaptation in the temporal, spatial or quality domain. Instead it reuses cached video frames if the transmission of current frames is not possible. The advantage of reusing cached frames is that the real-time aspect can be maintained for small time periods without sending a single byte of video data over the network. In case of short network failures it is not even possible to realize the difference between the original video and the reused previous frames. We evaluate and compare our frame reusing approach against adaptation in the quality domain which is a well know, but highly CPU intensive and non scalable technique.

Semantic Based Error Avoidance and Correction for Video Streaming

Video streaming over best effort networks remains a challenging task. Video quality decreases with an increasing number of frames that are corrupted, lost or only received after playback time. We use semantic information about the video and the network to decide between alternative or cooperative streaming sources to avoid or to correct data loss. We propose a distributed architecture that combines a peer-to-peer indexing archive for videos with error avoidance and error correction mechanisms to select the best delivery method from the corresponding sources. Our indexing-cache peer-to-peer overlay has two interesting properties for our selection model: it efficiently locates several sources for a video (if they exist) and even rare videos. Based on the coding characteristics of the available videos and the state of the network we apply a model for selecting between error avoidance, error correction and a combination of both approaches. This model is evaluated by using the network simulator NS-2 and a modified version of EvalVid.

A demo for affinity based content delivery architectures

Streaming video data over best effort networks is a challenging task concerning the quality of the received content. The quality decreases with the number of frames that are corrupted, lost or received after the playback time. The main reasons for lost, delayed or corrupted frames are overloaded streaming servers and crowded network paths. In order to deal with overloaded streaming servers and crowded network paths, we present the prototype of an innovative architecture called Proxy-to-Proxy (X2X). The three main components in the architecture are proxies (provided by end users), videos (replicated from original servers) and end-clients (requesting shared content). The behavior of the three components is defined by the so-called affinity model which consists of three levels. The first level covers the content replication from original to surrogate servers (also called proxies), the second level covers the collaboration between the proxies, and the third level covers the delivery from the proxies to the end-clients. Proxies, videos and end-clients have a certain affinity to each other. Only by changing the input parameters to the affinity functions, the overlay network behaves either like a classical 1) Content Delivery Network, 2) a Peer-to-Peer System or 3) a Proxy-to-Proxy overlay network --- which is a combination and generalization of the two former ones.