The ODDness of Webpages

Abstract

Users and content providers want websites to load quickly. A widely used web performance metric that rewards both the early appearance of content and the timely completion of page load is the Speed Index (SI). Lower SI values correspond to higher user satisfaction, which makes reducing page SI an important goal.In this paper, we observe that all images on a webpage are not created equal and indeed vary considerably along a metric we dub image density, or the ratio of byte weight to pixel size. Variation in image density creates opportunities to prioritize lower density images to reduce page SI by displaying more pixels sooner for every loaded byte. We define Object Density Distribution (ODD) – a new webpage characterization metric. To understand the potential for image prioritization, we characterize ODD of existing webpages, their ODDness if you will, and show that ODD skewness and kurtosis indicate meaningful prioritization opportunities. To understand the effectiveness of image prioritization, we propose a URL-based prioritization mechanism and measure its performance across 20 test pages loaded from the Apache, NGINX, and Caddy servers. Our results show SI improvement over 40% in some cases and mean improvement of 5.7%. These SI improvements and the simplicity of our prioritization method create a compelling case for the adoption of our method by content distribution networks (CDNs) and future browser implementations.