Data at Cloudflare scale: some insights on measurement for 1,111 interns

Cloudflare recently announced its goal to hire 1,111 interns in 2026, a figure that equates to about 25% of its full-time workforce and promises opportunities to develop and ship code into production. The piece explains why measuring data at such scale is both challenging and fruitful, noting that Cloudflare processes data from more than 45 million HTTP requests per second across 285 data centres and also handles TCP connections to its 1.1.1[.]1 DNS server.

Ram Sundara Raman’s internship project focused on detecting and explaining connection tampering by middleboxes, using a passive approach with production data and 19 tampering signatures, and was published live on Cloudflare Radar. The article discusses the limits of passive data, such as not knowing who tampered or exactly which domains were blocked, and highlights the need for a combination of passive and active probing to paint a fuller picture of Internet tampering.

It also recounts the Iran nationwide tampering signals observed in late 2022 and emphasises that large-scale observation requires careful handling of noise, sampling, and ground truth. According to Cloudflare Radar, the work demonstrates that even with vast data, observers must carefully sift signal from noise to derive meaningful insights.