Cloudflare通过数学优化和Rust改进节省100TB内存

Cloudflare通过对一致性哈希算法的数学优化和代码改进，成功将其Pingora后端路由服务的内存占用大幅降低1。该公司在pingora-ketama一致性哈希库中实施了多项优化措施，最终在全球范围内节省了超过100TB的内存1。

具体优化方案包括两个关键改进1。首先，Cloudflare将数据结构中的Point结构从8字节压缩至6字节，实现了存储空间25%的减少1。其次，通过数学推导，该团队证明可以将每台服务器的哈希值数量从160个减少至16个，在不影响误差率的前提下大幅降低计算开销1。优化后的新版本通过分层迁移策略在全球数据中心安全部署1。

Cloudflare has successfully reduced memory consumption across its Pingora backend routing service by over 100TB through mathematical refinements to its consistent hashing algorithm and code improvements written in Rust. 1 The optimization initiative focused on the pingora-ketama consistent hashing library, which had become a significant source of memory overhead. 1

The company implemented two primary optimization strategies. First, it compressed the Point data structure from 8 bytes to 6 bytes, achieving a 25% reduction in storage space. 1 Second, through mathematical analysis, Cloudflare demonstrated that the number of hash values per server could be reduced from 160 to 16 without degrading accuracy or error rates. 1 These improvements were subsequently deployed globally across Cloudflare's data centers using a phased migration approach to ensure safe rollout. 1

This memory optimization effort builds on similar work completed by Cloudflare's DNS team, which had separately achieved 100TB in memory savings the previous month. 1