Metraipolis › archive › dc_032bd04174fd

GraniKV: Asymmetric Granularity KV-Cache Paging for Multi-Agent Systems with Long Shared Prefix — arXiv:2608.15584

Source: arXiv (en)
Original: https://arxiv.org/abs/2608.15584
License (as archived): CC0-1.0
Archived: 2026-10-18T18:03:48.840Z
SHA-256 of the archived text: 97d8d131da1fedf0c5c3b16cb8d65f917ca045633ef3cc6663073ca65a821dfe
Stable URL of this page: https://metraipolis.com/documento/dc_032bd04174fd

How to cite

GraniKV: Asymmetric Granularity KV-Cache Paging for Multi-Agent Systems with Long Shared Prefix — arXiv:2608.15584 — arXiv (en), https://arxiv.org/abs/2608.15584 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_032bd04174fd (Metraipolis archive dc_032bd04174fd, sha256 97d8d131da1fedf0c5c3b16cb8d65f917ca045633ef3cc6663073ca65a821dfe)

BibTeX

@misc{dc032bd04174fd,
  title = {GraniKV: Asymmetric Granularity KV-Cache Paging for Multi-Agent Systems with Long Shared Prefix — arXiv:2608.15584},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2608.15584},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_032bd04174fd; sha256 97d8d131da1fedf0c5c3b16cb8d65f917ca045633ef3cc6663073ca65a821dfe},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.15584. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

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 $\mathbf{2.16\times}$, $\mathbf{1.98\times}$, and $\mathbf{1.57\times}$ 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 $\mathbf{1.95\times}$ while batch-global cascade collapses to parity --- the storage layer alone carries the win in the regime that motivates the paper.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-08-18 in cs.AI).
Categorie arXiv: Machine Learning (cs.LG) ; Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2608.15584 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-18T16:29:14.029Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-08-18?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).