Impactos da interatividade na escalabilidade de protocolos para mídia contínua

Abstract

A number of scalable bandwidth sharing protocols have been proposed, with the goal of greatly reducing the server and network bandwidth required to deliver popular media. The scalability of these protocols have been assessed for sequential workloads. However, high degree of interactivity has been observed in real media workloads. Furthermore, a number of studies show that interactivity seriously affects the scalability of the protocols and, thus, can compromise the popularity of streaming media in the Internet. In this area, immediate service protocols are of particular interest, as they serve clients with no or little delay, and can be used where fast interactive response is required.We investigate the impact of interactivity in the scalability of bandwidth sharing streaming media protocols with immediate service, and propose new strategies do reduce this impact. We identify a reduced set of interactive media workloads with impact in the scalability of such protocols, and use these factors as a basis to create a model for streaming media workloads. We use this model, together with a rich set of real streaming media workloads, to build a rich set synthetic interactive workloads which covers a wide range of interactivity profiles. We use this set of workloads to evaluate the impact of interactivity in 'Patching' and 'Bandwidth Skimming'.A number of different and complementary strategies is proposed to reduce the impact of the interactivity of 'Bandwidth Skimming', which scales better than 'Patching' for sequential and interactive workloads.We uncover the degenerative effect of 'request fragmentation' which, in some degree, affects any protocol which uses client-side buffering to reduce data retransmission. In order to comprehensively cover the design space of streaming media protocols, we also propose hybrid strategies which combine efforts of different individual strategies.'Dispersion' is defined as a factor of streaming media workloads which is key to the scalability of protocols. We start with a qualitative approach to dispersion, which is used as a framework for protocols analysis and evaluation. We follow with a quantitative approach to dispersion, which is used to estimate the server bandwidth required to the Bandwidth Skimming protocol.As any strategy is affected by the degenerative effect of request fragmentation, we propose an 'adaptive protocol'. This protocol monitors the workload, measuring it's dispersion, and uses this dispersion as a guide to decide which strategy, if any, to apply to a given workload. This protocol can be successfully used with any of our workloads (educational video, entertainment video and audio), and avoids the degenerative effect of request fragmentation. This protocol saves up to 66% of server bandwidth if compared to the original Bandwidth Skimming protocol.