Aries: Rethinking AI Cloud Infrastructure for Agentic Serving Systems

PROBLEMA / La señal entra en el radar porque La infraestructura LLM actual optimiza tokens y batches, pero los agentes alternan razonamiento, llamadas externas, esperas y ráfagas cortas de ejecución. Por qué puede ser importante: Muestra que las métricas token-centric ocultan cuellos de botella relevantes; que añadir más contexto produce rendimientos decrecientes; y que los sandboxes permanecen ociosos durante largos intervalos, aunque suspenderlos resulte caro.

MÉTODO / La lectura de 2. Background and Motivation describe la intervención y su construcción: We first discuss the differences between the conventional LLM serving and emerging agentic workloads, and their implications for modern cloud AI infrastructure. We then discuss the existing benchmarking frameworks and why they fall short for systems research in agentic systems. Conventional LLM serving systems are organized around the execution of individual model requests. As shown in Fig. 1 a, the serving runtime processes a prompt, generates output tokens, manages accelerator memory, and returns a response ( 53 ; 50 ) . Scheduling and memory management are therefore scoped primarily to the active invocation.… [Fuente: https://arxiv.org/html/2607.29069#S2]

RESULTADO / La sección 4. Why Is Modern Cloud AI Infrastructure a Poor Fit for Agents? informa: We sample 20 tasks from each benchmark to achieve 50% task success rate (important for the accuracy studies) with OpenClaw and repeat each task five times. Retaining excessive history can increase tail delay, while aggressive compression may discard information required by later steps and reduce task accuracy ( 20 ; 21 ) . In the latter analysis, across all evaluated benchmarks, we sweep the context budget for both OpenClaw and Hermes Agent (omitting OpenHands for brevity) to determine when additional retained history stops providing meaningful accuracy gains. [Fuente: https://arxiv.org/html/2607.29069#S4]

LÍMITE / El cierre de la fuente señala: Agent serving turns conventional LLM serving into long-lived, stateful trajectories spanning model execution, harness logic, and sandboxes. We present Aries, a telemetry framework that combines reproducible experiments with production traces to characterize task and system behavior. Our findings motivate trajectory-capturing metrics, control planes for context management and agent-aware elasticity, and sandboxes with minimal attack surfaces. La transferencia a cloud agentic requiere repetir la comparación con datos y criterios propios [Fuente: https://arxiv.org/html/2607.29069#S6].