An Experimental Evaluation of Google’s QUIC Protocol

Vamshi Reddy Konagari
Anise Ghorbani

ABSTRACT

Traditional transport protocols for the web are not suitable for the
multiplexed stream delivery requirements of HTTP/2. This is due to
problems such as head of line blocking between streams and high
connection establishment latency. QUIC is an experimental protocol
developed by Google to address these shortcomings. QUIC runs on UDP
with 2-RTT initial connection establishment and 0-RTT subsequent
connection establishment, reliable delivery of data, stream
multiplexing, cubic congestion control, and more. Through extensive
experimentation, this paper investigates and compares the performance
of QUIC to HTTP/1 and HTTP/2. It was found that QUIC is able to
perform better than other protocols in conditions of high packet loss,
high latency, and low bandwidth. Also, QUIC shares links fairly with
other traffic flows, while maintaining high link utilization. Finally,
several transport-layer performance issues encountered in Long Fat
Networks are revealed, when using either TCP or QUIC.