Abstract

In order to minimize the downloading time of short-lived applications like Web browsing, Web application, and short video clips, the recently standardized HTTP/2 adopts stream multiplexing on one single TCP connection. However, aggregating all content objects within one single connection suffers from the head-of-line blocking issue. Quick UDP Internet connection (QUIC), by eliminating such an issue on the basis of UDP, is expected to further reduce the content downloading time. However, in mobile network environments, the single connection strategy still leads to a degraded and high-variant completion time due to the unexpected hindrance of congestion window growth caused by the common but uncertain fluctuations in round trip time and also random loss event at the air interface. To retain resilient congestion window against such network fluctuations, we propose an intelligent connection management scheme based on QUIC which not only employs adaptively multiple connections but also conducts a tailored state and congestion window synchronization between these parallel connections upon the detection of network fluctuation events. According to the performance evaluation results obtained from an LTE-A/Wi-Fi testing network, the proposed multiple QUIC scheme can effectively overcome the limitations of different congestion control algorithms (e.g., the loss-based new reno/CUBIC and the rate-based BBR), achieving substantial performance improvement in both median (up to 59.1%) and 95th completion time (up to 72.3%). The significance of this piece of work is to achieve highly robust short-lived content downloading performance against various uncertainties of network conditions as well as with different congestion control schemes.

Highlights

  • Nowadays, as the content volume downloaded through mobile wireless networks have already exceeded their wired counterparts, downloading acceleration of the diverse shortlived applications like webpage browsing, web application and short video clips in such environments has attracted significant research attentions in both academia and industry [1]

  • The inherited Congestion Control Algorithms (CCA) on the single connection of Quick UDP Internet Connection (QUIC) still suffer from unexpected hindrance of congestion window growth caused by the unavoidable network fluctuation events (NFE) such as packet loss and round-trip time (RTT) variations which are very common in wireless environments

  • In this paper, we proposed an mQUIC scheme that employs multiple connections instead of the default single UDP connection adopted by QUIC furthering order to comprehensively enhance the performance of web content with QUIC in mobile networks

Read more

Summary

INTRODUCTION

As the content volume downloaded through mobile wireless networks have already exceeded their wired counterparts, downloading acceleration of the diverse shortlived applications like webpage browsing, web application and short video clips in such environments has attracted significant research attentions in both academia and industry [1]. Compare to HTTP/2 over TCP, the content downloading time of QUIC is substantially improved in lossy network environments [9], and the reduced handshake complexity and flexible stream multiplexing yield notable efficiency improvement in the delivery of today’s short-lived applications [9] Despite these improvements, the inherited Congestion Control Algorithms (CCA) on the single connection of QUIC still suffer from unexpected hindrance of congestion window growth caused by the unavoidable network fluctuation events (NFE) such as packet loss and RTT variations which are very common in wireless environments. For rate-based CCA like BBR, mQUIC can successfully speed up its median and 95th downloading time up to 27.4% and 31.4 %, respectively This technique is able to achieve highly robust content downloading performance under various network conditions, content size, as well as the initial congestion window size. This is in contrast to the plain QUIC-based approach where the actual performance can be very sensitive to specific network conditions and configurations

AND RELATED WORKS
OPTIMIZING LOSS-BASED CCAs
Findings
CONCLUSION
Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call