Leveraging fast I/O of unikernel in QUIC protocol

Abstract

Beyond its inception as a toy project at Google, QUIC has emerged as a capable replacement for the traditional TCP/HTTPS protocol stack. Due to its manifold benefits including latency reduction and the elimination of HOL (Head of Line) blocking, QUIC is now estimated to account for at least 9.1% of the Internet traffic [3]. However, there has been a concern with its high CPU consumption as it is placed in user space to avoid entrenchment. In this thesis, we examine a Unikernel approach to alleviate QUIC’s higher CPU usage with I/O. Our two prototypes show reduced CPU time in the I/O of QUIC implementations—ranging from 1.29% to 35.94% depending on scenarios—while the reduction is offset by increased CPU time in other parts.