Accelerating mobile web downloading through context-aware and cross-layer intelligence

• Main Author: Qian, Peng
• Other Authors: Wang, Ning ; Tafazolli, Rahim
• Published: University of Surrey 2018
• Subjects: 621.3
• Online Access: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.766985

Web-based content is a dominant application type in mobile network but accessing such content suffers from poor downloading latency. In modern mobile networks, accelerating web content downloading faces three distinctive challenges. First the web contents enter a rich-media era, with an explosion of the content size and an evolution of content structure which not only requires increased network resources but also incurs noticeable computation latency. Second the unavoidable network uncertainties like RTT variation and random loss aggravate such degraded downloading time, although the network has already offered augmented resources like high bandwidth, low packet loss and latency. Third, the newly standardised protocols like HTTP 2.0 and QUIC are expected to provide an optimised resource utilisation, but existing understanding of such protocols when applying on web content is still superficial. By realising these intertwined technical aspects, we examined three web downloading scenarios, figured out how these aspects qualitatively affect downloading time and then proposed optimisation intelligence accordingly. First, we focused on the fixed single connection number of HTTP 2.0 which cannot be adaptive for various content size and network conditions. By clarifying the numerical relationship between content size, network condition and connection number, we proposed a context-aware mobile edge hint framework. In this framework, a mobile edge hint server offline collects the meta-data of popular webpages as well as the network condition and performs online hints of such information upon receiving the user request. Then the user can execute a novel algorithm to select an optimal connection number by understanding the specific network condition and content characteristics through the edge hint. Both numerical and test-bed based results validate that this framework can bring a noticeable acceleration of webpage downloading. Second, we turned our attention to the computation latency which is caused by the unavoidable computation task during webpage downloading. We seek for a transport layer approach since pure application layer approaches are recognised to have practicality and security limitation. To this end, a non-URL based mobile edge computing framework is proposed to serve a novel transport layer IW selection algorithm at the client side. This framework is validated to have remarkable performance improvement when computation latency occupies less than 50% of total downloading time. Third, we investigated QUIC's performance on web content, especially in the presence of network uncertainties. The evaluation results carried out on real mobile networks reveal that the different congestion control algorithms plugged in QUIC can lead to distinctive shortages under network fluctuations. Then we proposed a mQUIC scheme which performs a customised state and congestion window synchronisation algorithm based on multiple coordinated connections. We conducted extensive evaluations of mQUIC and the results substantiated faster and robust downloading time can be achieved by mQUIC when compared to plain QUIC enable contents.