In this dissertation, we study the problem of effective traffic management in the heavily loaded networks for videoconferencing from the real-time stochastic system perspective. A probabilistic delay model is proposed for the characterization of time-varying nature of the data traffic in the Internet. Connectionist delay representation of the network traffic is derived by considering a videoconference session among a group of Nparticipants connected via the Internet. Connections among the collaborative participants are represented by their end-to-end network delays statistics involving the characteristics of end-to-end transmission delay of each connection and the inter-connection relation. The first component is represented by the probability density function of the end-to-end delay in each connection. The second component is modeled as the delay correlation between the connections. For a network of Nworkstations, this model becomes an N × Nmatrix of probability density functions and N 2× N 2correlation matrix that, when normalized by the delay mean and variances of individual connections, leads to matrix of delay cross-correlation coefficients with values varying in the range of [-1, 1] independent of scale changes in the correlation amplitude.
For two connections, the higher the value of the cross-correlation coefficient the stronger the correlation between them is. This also means that two connections are sharing a portion of the Internet for their own communications. The shared portion of the network is the physical link that causes the delay due to the heavy traffic for the two connections in question. This also suggests that to reduce the network traffic through the shared link the media data send only once to one workstation and then passed to the second one. We can accomplish this management strategy by clustering a set of participants that their connections are highly correlated to other into groups and assigning a local manager to each group. Using this strategy, a hierarchical management is established among the collaborating participants so that each group has minimal effect on each other.
Using the proposed end-to-end delay model, the packet loss rates associated with each connection can be estimated. For minimum packet loss rate, a scheduling algorithm is proposed that determine the time interval that has minimum packet loss rate. The scheduling algorithm searches linearly the time for optimal time that has a minimum packet loss rate for N 4connections.