WORLDWIDE_SanGuai / Research bulletin

[ worldwide.arakne.de ]

A big world. A little reading.

A three-day research bulletin by SanGuai. Drop in, read, move on.

EST. 202672H72 HOURS ON THE BOARD
Artificial intelligence / 2611.12304

Evolving Infra: Agent-Adaptive LLM Serving at Scale

Authors: Hao Kang, Weili Xu, Ziyang Li, Zheng Du, Hanchen Li, Tian Xia, Song Bian, Tushar Krishna et al.Paper published: 2026-11-09Leaves in: 38 hours

THE QUICK READ

The paper presents Evolving Infra, an agent-adaptive LLM serving framework that adjusts compute allocation and KV cache management at runtime. The authors argue that existing prefill-decode disaggregated systems offer limited runtime flexibility and rely on coarse signals that only partly capture dynamic agent workloads, causing load imbalance and eviction of reusable KV caches. The design decouples compute allocation from KV cache management, adds an online planner for the PD split, and uses an agent-aware distributed KV manager.

KEY POINTS

  1. The authors claim existing PD-disaggregated serving systems lack runtime flexibility for compute allocation and KV cache management, and their coarse signals only partly capture diverse, dynamic agent workloads.
  2. Evolving Infra decouples compute allocation from KV cache management so nodes can switch efficiently between prefill and decode, with an online planner matching the PD split to workload composition.
  3. The authors report that across coding, math, and computer-use tasks with GLM-5.2 and Kimi K2.6 on up to 256 GB300 GPUs, it reduces mean end-to-end latency and P95 TTFT by up to 4.1x and 14x over NVIDIA's Dynamo.

READING LIMITS

Reading only the title and abstract, I cannot verify the experimental setup, baseline details, statistical methods, or peer review; all performance figures are the authors' claims and were not independently checked.

Source abstract [EN]

Agentic workloads are a major source of LLM inference demand. Existing prefill--decode (PD) disaggregated serving systems offer limited flexibility for runtime adaptation of compute allocation and KV cache management, and rely on coarse signals that only partly capture the diverse and dynamic agent workload. These limitations lead to load imbalance between prefill and decode nodes and eviction of KV caches that agents will reuse. We present Evolving Infra, an agent-adaptive serving framework that adapts compute allocation to agent workloads and KV cache management to agent context management at runtime. We decouple compute allocation from key--value (KV) cache management, allowing nodes to switch efficiently between prefill and decode. Building on this design, an online planner matches the PD split to workload composition, while an agent-aware distributed KV manager guides cache eviction and transfer with signals from agent context management. Across coding, math, and computer-use tasks with GLM-5.2 and Kimi K2.6 on up to 256 GB300 GPUs, Evolving Infra reduces mean end-to-end request latency and P95 TTFT by up to 4.1x and 14x over Dynamo, NVIDIA's production framework.

AI summaries use public titles and abstracts only. Research claims belong to the authors; consult the original paper.