Cloudflare reclaims 100TB of RAM by reworking consistent-hashing point counts

Sources: Cloudflare blog post, HN discussion
Summary: The saving lands in Pingora Backend Router, through the pingora-ketama consistent-hashing library. The post derives the coefficient of variation for ketama ring segments as a function of points per server, and shows the 160-points-per-server default inherited from NGINX is what drives the memory cost. Cloudflare reports more than 100TB reclaimed here, on top of a separate 100TB the DNS team shed the previous month.
Why it matters: The derivation is reusable by anyone running consistent hashing at a scale where ring size dominates memory.