TokenDance: Scaling Multi-Agent LLM Serving via Collective KV Cache Sharing

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Bian, Zhuohang, Wu, Feiyang, Zhang, Chengrui, Dong, Hangcheng, Liang, Yun, Zhuo, Youwei
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911565310066688
author Bian, Zhuohang
Wu, Feiyang
Zhang, Chengrui
Dong, Hangcheng
Liang, Yun
Zhuo, Youwei
author_facet Bian, Zhuohang
Wu, Feiyang
Zhang, Chengrui
Dong, Hangcheng
Liang, Yun
Zhuo, Youwei
contents Multi-agent LLM applications organize execution in synchronized rounds where a central scheduler gathers outputs from all agents and redistributes the combined context. This All-Gather communication pattern creates massive KV Cache redundancy, because every agent's prompt contains the same shared output blocks, yet existing reuse methods fail to exploit it efficiently. We present TokenDance, a system that scales the number of concurrent agents by exploiting the All-Gather pattern for collective KV Cache sharing. TokenDance's KV Collector performs KV Cache reuse over the full round in one collective step, so the cost of reusing a shared block is paid once regardless of agent count. Its Diff-Aware Storage encodes sibling caches as block-sparse diffs against a single master copy, achieving 11-17x compression on representative workloads. Evaluation on GenerativeAgents and AgentSociety shows that TokenDance supports up to 2.7x more concurrent agents than vLLM with prefix caching under SLO requirement, reduces per-agent KV Cache storage by up to 17.5x, and achieves up to 1.9x prefill speedup over per-request position-independent caching.
format Preprint
id arxiv_https___arxiv_org_abs_2604_03143
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle TokenDance: Scaling Multi-Agent LLM Serving via Collective KV Cache Sharing
Bian, Zhuohang
Wu, Feiyang
Zhang, Chengrui
Dong, Hangcheng
Liang, Yun
Zhuo, Youwei
Distributed, Parallel, and Cluster Computing
Multi-agent LLM applications organize execution in synchronized rounds where a central scheduler gathers outputs from all agents and redistributes the combined context. This All-Gather communication pattern creates massive KV Cache redundancy, because every agent's prompt contains the same shared output blocks, yet existing reuse methods fail to exploit it efficiently. We present TokenDance, a system that scales the number of concurrent agents by exploiting the All-Gather pattern for collective KV Cache sharing. TokenDance's KV Collector performs KV Cache reuse over the full round in one collective step, so the cost of reusing a shared block is paid once regardless of agent count. Its Diff-Aware Storage encodes sibling caches as block-sparse diffs against a single master copy, achieving 11-17x compression on representative workloads. Evaluation on GenerativeAgents and AgentSociety shows that TokenDance supports up to 2.7x more concurrent agents than vLLM with prefix caching under SLO requirement, reduces per-agent KV Cache storage by up to 17.5x, and achieves up to 1.9x prefill speedup over per-request position-independent caching.
title TokenDance: Scaling Multi-Agent LLM Serving via Collective KV Cache Sharing
topic Distributed, Parallel, and Cluster Computing
url https://arxiv.org/abs/2604.03143