REWINDING THE EXPONENTIAL: HOW WE KEEP HARDWARE USAGE DOWN

Summary from the TLDR Data newsletter

Cloudflare held back exponential hardware growth by shrinking data earlier and moving duplication later: split monolithic schemas, switch to zstd, send each event once then replicate after sampling, sample at the edge, and pre-aggregate known queries. Optimize bytes, not just compute: prune fields, compress measured workloads, push reductions upstream, and replicate compacted data.