ScaleSim: Serving Large-Scale Multi-Agent Simulation with Invocation Distance-Based Memory Management
This solves memory bottlenecks for researchers and developers running large-scale multi-agent simulations, though it is incremental as it builds on existing serving systems.
The paper tackles the problem of scaling LLM-based multi-agent simulations by addressing GPU memory exhaustion from agent-specific states, introducing ScaleSim with invocation distance-based memory management that achieves up to 1.74x speedup over SGLang on benchmarks.
LLM-based multi-agent simulations are increasingly adopted across application domains, but remain difficult to scale due to GPU memory pressure. Each agent maintains private GPU-resident states, including models, prefix caches, and adapters, which quickly exhaust device memory as the agent count grows. We identify two key properties of these workloads: sparse agent activation and an estimable agent invocation order. Based on an analysis of representative workload classes, we introduce invocation distance, a unified abstraction that estimates the relative order in which agents will issue future LLM requests. Leveraging this abstraction, we present ScaleSim, a memory-efficient LLM serving system for large-scale multi-agent simulations. ScaleSim enables proactive prefetching and priority-based eviction, supports diverse agent-specific memory through a modular interface, and achieves up to 1.74x speedup over SGLang on simulation benchmarks.