Cloudflare Introduces Meerkat — A Global Consensus Experiment

Cloudflare Unveils Meerkat — An Experiment in Global Consensus

Cloudflare’s research team has published Meerkat, a new globally-distributed consensus protocol designed to tolerate slow or adversarial network paths while keeping throughput high. The prototype runs across Cloudflare’s existing edge fleet and is positioned as research-stage, not a product.

What’s new

• A leaderless consensus design that does not stall under partial WAN partitions
• Demonstrated throughput in the tens of thousands of small operations per second per region
• Implementation intentionally lightweight — written to fit inside a single Worker isolate

Why it matters

• Today’s internet-scale coordination (KV, queues, durable objects) leans on per-region or per-pod consensus — Meerkat explores truly global
• If it lands, it could collapse a class of “control-plane sharding” decisions inside the edge platform
• Cloudflare has a track record of research-to-product (e.g. Workers, R2, D1) — this is worth tracking

Caveats

• No SLA, no GA timeline — this is a paper and a prototype, not a service
• Adversarial model assumes honest majority of Cloudflare’s own edge nodes