Reducing HTML cache footprint through tracking data optimization: Insights from eBay

Background Study: Hypertext Markup Language (HTML) caching [1] is a practice used in software systems to serve data quicker to incoming Hypertext Transfer Protocol (HTTP) requests [2]. Systems may cache rendered HTML or pre-rendered HTML either within the system or on the edge servers. Modern HTML renderer systems like React [3] and Marko [4] incorporate heuristics in the form of comments and attributes within the rendered HTML to facilitate efficient render and re-render processes.Problem Statement: In addition to these heuristics from rendering systems, the associated tracking systems embed additional data payload into HTML for end-user behavior tracking [5] These data payloads are crucial for internal tracking systems to assess the effectiveness of user features, based on the user behavior. However, these tracking data heuristics significantly increase the size of the HTML, posing challenges in scenarios where caching of the said HTML is required. The bulkiness of tracking data heuristics compared to rendering heuristics leads to inefficiencies in HTML caching.Proposed Model/Technique: This paper proposes an optimization technique that is focused on minimizing the size of the tracking data heuristics embedded in HTML. The technique aims to balance the need for effective user tracking with the efficiency of HTML caching. The novelty of the approach is that, it does not do pre-processing or post-processing of the embedded tracking heuristics but performs the said optimization alongside the render of the HTML, during the computation of the hydration [6] data. The approach involves a case study of eBay’s systems, where we explore methods to refine the data payloads without compromising the tracking capabilities.Results: The implementation of the proposed optimization technique in eBay’s systems demonstrated a significant reduction in cached HTML sizes, both in edge caching [7] and on the application servers. The optimized tracking heuristics maintained their efficacy in tracking user interactions and user behavior for specific product features, therefore showcasing the potential of this methodology to enhance caching efficiency without losing critical tracking functionalities and capabilities.