Reproduce “KVCompass: A Discovery Plane for Internet-Scale Prefix Cache Sharing in LLM Serving” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Shiyi Ling, Fei Fang, Chen Qian
KVCompass introduces a discovery plane for cross-region LLM prefix-cache reuse, choosing workers by network distance, queueing, remaining prefill, and cache tier. Local indexes track cache locations, while the router scores all workers, including those that must recompute. In a 64-worker, eight-region simulation, P99 TTFT on Mixed at 3,000 requests/s was 14× lower than with hash-placement schemes—a simulation, not production.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.