Augmented Backpressure for Decentralized Management of Agentic Networks

2026-08-01

AI Summary

Traditional queue control methods struggle to handle the dynamic load and task dependencies arising from contextual memory states—such as KV caches and retrieved contexts—in agent networks, leading to suboptimal performance. This work proposes the Memory-Augmented Backpressure (MABP) framework, which, for the first time, incorporates contextual memory into queue control decisions. MABP jointly models request queues and causal memory dynamics, leveraging a memory-dependent pressure metric to co-optimize routing, transmission, activation, and service actions at each time slot while preserving effective memory under budget constraints. Theoretical analysis demonstrates that MABP strictly enlarges the system’s stability region and achieves throughput optimality with drift-plus-penalty guarantees: exact solutions attain a tight capacity bound, while approximate solutions incur only controllable performance loss.

Abstract

Agentic foundation-model service networks handle requests spanning retrieval, planning, generation, verification, and tool use. Unlike traditional communication networks, control performance depends on queue dynamics and contextual memory states, including prefix/KV blocks, retrieved contexts, expert warm states, and verified tool outputs. These states arise from execution history and alter service work and downstream successor laws under finite local budgets. Treating them as passive caches or an independent process leaves a queueing-control gap. To this end, we propose \emph{Memory-Augmented Backpressure} (MABP), a queue--memory control framework for stateful foundation-model service networks (SFMSNs) that jointly models commodity queues and causal contextual memory dynamics. MABP represents each request by service and state types, estimates memory-dependent work, penalties, and successor probabilities, then reads queues and resident memory each slot, selects feasible routing, transfer, activation, and service actions using a memory-dependent pressure score, and retains a budget-feasible subset of resident and newly generated objects. We prove an occupation-measure capacity outer bound with conditional tightness. We show that modeling contextual memory can strictly increase the stability region through work reduction and transition shaping, establishing a separation between memory-aware and memory-oblivious decisions. We also prove throughput and drift-plus-penalty guarantees for exact frame-MABP with bounded-loss extensions to approximate solvers.

Problem

• Agentic Networks
• Contextual Memory
• Queueing Control
• Stateful Service Networks
• Backpressure

Innovation

• Memory-Augmented Backpressure
• stateful foundation-model service networks
• contextual memory dynamics
• queueing control
• stability region