What Facebook’s New Study Reveals About Web Cache Effectiveness in 2024

Translator's Preface

Eight years ago Yahoo conducted a detailed study of web caching. With the rapid evolution of the Internet, Facebook's web team has now collected and analyzed fresh data on cache behavior for both desktop and mobile resources.

Main Text

Page load speed is crucial for every website, yet caching—allowing browsers to reuse previously downloaded resources—is often overlooked despite its significant impact on performance.

Re-Research

Inspired by Yahoo’s methodology, Facebook created a PHP endpoint that serves an image while logging each request in a database. The image’s HTTP headers control browser and proxy caching. The following diagram shows the original header configuration:

Due to known bugs in IE7 and IE8, the two header attributes were replaced as shown below:

When a browser requests the image, two scenarios occur:

If the image has never been loaded, the server returns a 200 Success response with the image data, and the browser caches the Last-Modified and ETag values.

If the image was previously loaded, the browser sends If-None-Match or If-Modified-Since headers; the server can respond with 304 Not Modified, avoiding retransmission of the image data.

The image was embedded in Facebook’s search bar so that each page load would trigger the caching test.

Research Results

After several weeks of data collection, analysis of the most recent seven days showed that 25.5% of requests still resulted in an empty cache. Breaking the data down, 24.8% of desktop requests and 26.9% of mobile requests were uncached.

Further segmentation of desktop browsers revealed higher cache rates for Chrome and Opera. Firefox data was excluded because versions ≤31 showed an 80% cache rate, while newer versions conflicted with our measurement method.

Mobile data showed cache rates ranging from 68% to 84%, with variations likely due to lower‑end devices.

The proportion of empty‑cache users for mobile versus desktop is illustrated below:

Overall, 44.6% of users had an empty cache, closely matching Yahoo’s 2007 findings.

Further

Given Facebook’s rapid release cycle (two versions per day), we investigated optimal cache lifetimes by measuring the time between the first successful request and the first 304 response. The resulting chart shows the distribution of cache acquisition times.

The horizontal axis represents hours; the vertical lines P50 and P75 indicate the time by which 50% and 75% of requests have cached resources. For mobile users, 50% of requests are cached after roughly 12 hours.

Practical Application

Our findings align with the 2007 study: when excluding newer Firefox versions, the maximum cache hit rate reaches 84.1%, surpassing the earlier 80% peak. However, cache lifetimes remain short—about 47 hours after a new version release, only 42% of requests are cached.

The growth of web page sizes (from ~168 KB in 2007 to >1 MB today) stresses browser caches, making efficient caching more critical than ever.

Best practices include using external CSS/JS, setting Cache‑Control and ETag headers, compressing assets, employing versioned URLs, and separating frequently updated resources. These recommendations benefit large‑scale sites like Facebook as well as smaller websites.