GraniKV: Asymmetric Granularity KV-Cache Paging for Multi-Agent Systems with Long Shared Prefix

Abstract

Production paged-serving engines apply uniform paging granularity to the KV cache, even though the two regions of a multi-agent workload have opposite storage requirements: a long shared prefix demands contiguity, while the per-request suffix demands fine-grained allocation. We present \textbf{GraniKV}, a KV-cache layer that allocates the shared prefix in a contiguous HOT pool and the suffix in a token-level COLD pool, combined with a per-step dispatcher which selects the appropriate backend among dual backends for each regime (compute-, memory-, or communication-bound). To the best of our knowledge, GraniKV is the first system to apply asymmetric paging granularity to the KV cache of a production paged-serving engine. At $L_p=16$\,K shared tokens GraniKV reaches $2.16\times$, $1.98\times$, and $1.57\times$ output-token throughput over the production baseline on Llama-3.1-8B/TP=1, Qwen-2.5-14B/TP=2, and Qwen-2.5-32B/TP=4. The gain decomposes: cascade attention integration contributes the majority at saturation; the asymmetric storage layer adds $1.05$--$1.15\times$ end-to-end while being what makes the batched-GEMM prefix backend possible at all. Under heterogeneous multi-agent serving with \emph{distinct} prompts of different lengths, the attribution inverts: GraniKV sustains $1.95\times$ while batch-global cascade collapses to parity --- the storage layer alone carries the win in the regime that motivates the paper.

An error in throughput measurement methods was found. The paper's details are deprecated