Methodology of improving video conference reliability

One of the main problems of organizing a reliable videoconference system is to provide minimum data transfer speed while maintaining maximum speed of audio and video stream processing. Existing software solutions have limita- tions on number of simultaneous participants in a videoconferencing session, hardware systems are expansive. Today the main methods for improving reliability of videoconferencing systems are methods of load sharing. The authors pro- pose a new technique for increasing the reliability of videoconferencing for authorized users with guaranteed messages delivery and model of probabilistic access to videoconference. The model contains two levels. The top-level model allows describing various videoconference systems taking into account their specific features. The low-level model describes videoconference systems for authorized users with guaranteed massage delivery. The model is based on the following concepts: subjects, objects, actions. To determine the probability of full access to a videoconference system consisting of several servers and clients it is necessary to determine the probability of their availability. The mathe- matical apparatus for evaluating the videoconference system for the standard mode is based on the multi-channel queue system with limited queue, and a set of homogeneous single-channel queue systems with limited queue for special mode. Evaluating of effectiveness of special and standard mode is performed by comparing the probability of access failure. For those calculations software developed by the authors is used. At the output software issues a graph in three coordinates P(ktotal, kspec), ktotal , kspec . From the graph for a given level P(ktotal, kspec) can be defined ktotal and kspec . As the exam- ple, the article contains the graph reflecting advisability of using a special mode in a videoconference system specified by numerical parameters. The graph shows the maximum number of special clients for predetermined total number of clients and given level of threshold probability P(ktotal, kspec).