How speedy is SPDY?


SPDY is increasingly being used as an enhancement to HTTP/1.1. Unfortunately, the performance of SPDY is not well understood. Both SPDY and HTTP performance depend on many factors external to the protocols themselves, including network parameters, TCP settings, and Web page characteristics. Any of these factors can have a large impact on performance, and to understand their interplay it is necessary to sweep a large portion of the parameter space.

By building on top of WProf, we present what we believe to be the most in-depth study of page load time under SPDY as of early 2014. We isolate the different factors that affect PLT with reproducible experiments that progress from simple but unrealistic transfers to full page loads. By looking at results across this progress, we can systematically isolate the impact of the contributing factors and identify when SPDY helps significantly and when it performs poorly compared to HTTP.

The following pages provide more details.

We release the data obtained by sweeping the parameter space. Further analysis on this data is highly welcome.
We release the tool that emulates the page load process for researchers and other folks who are only interested in running network experiments while hoping to preserve authenticity of the page load process. Unlike browsers, our tool performs deterministically in computation that saves lots of repeated experiments which statistically reduce the impact of variability in computation. By using our tool, it is easy to play with advanced SPDY mechanisms (prioritization and server push).
The technical paper that describes our experiments, results, and analysis.
Contact us for any related questions.