Автор(ы): Маркович Н. М. (ИПУ РАН, Лаборатория 38)Автор(ов): 1 Параметры публикацииТип публикации: Тезисы докладаНазвание: Risk Evaluation of Transmission Processes in Overlay NetworksНаименование конференции: 59th ISI World statistics congress 25-30 August Hong Kong, ChinaНаименование источника: Proceedings of the 59th ISI World statistics congress (Hong Kong, China, 2013)Город: Hong KongИздательство: City University of Hong KongГод издания: 2013Страницы: 8 АннотацияThe paper contributes to the risk evaluation and quality-of-service assessment of transmission processes in overlay networks. Peer-to-peer systems like Skype and Internet IP TV provide examples of overlay networks. The latter can be considered as multi-agent systems with a random number of players (active customers of Internet) at each time moment. Such randomness causes problems in the observation and gathering of statistical information regarding the transmission processes and the required resources for a stable operation of the network. Thus, inferences regarding distributions of the networks characteristics based on available information are required. We consider the packet transmission processes that are induced by a peer-to-peer overlay disseminating real-time traffic as an available example of such a multi-agent system. It is technically very difficult to measure the length of the packet path and the end-to-end delay (i.e. the packet delivery time between sender and receiver peers). The problem is also that inter-arrival times between packets are dependent and heavy-tailed distributed. This generates clusters of exceedances of the transmission rate process over sufficiently high thresholds. The latter are interpreted as equivalent capacity of the channel. The clusters cause the loss of packets due to exceedances over the available channel capacity. Applying the theory of extreme events, we derive that the asymptotic distribution of the end-to-end packet delay is subexponential heavy-tailed if the inter-arrival times between packets are regularly varying heavy-tailed distributed. The packet length of the path is derived to be geometrically distributed. Библиографическая ссылка: Маркович Н.М. Risk Evaluation of Transmission Processes in Overlay Networks / Proceedings of the 59th ISI World statistics congress (Hong Kong, China, 2013). Hong Kong: City University of Hong Kong, 2013. С. 8.