Fleet Sim Overview

Fleet Sim is the maintained fleet simulator for vLLM Semantic Router. The vllm-sr-sim package is its CLI and service entrypoint. It helps you plan GPU fleets before deployment, compare routing and split strategies, and expose those workflows inside the dashboard without reviving a separate simulator frontend.

What Fleet Sim is for

• sizing homogeneous, heterogeneous, or disaggregated fleets against a latency target
• comparing annualized cost across GPU choices, routing policies, and threshold choices
• validating planning assumptions with simulation runs, trace replay, and what-if analysis
• surfacing those workflows in the dashboard through a maintained backend proxy

What Fleet Sim is not for

• it is not the router's live request path
• it is not a runtime autoscaler or burst controller
• it is not a per-kernel profiler for one deployment replica
• it is not a replacement for the router configuration docs

Deployment modes

vllm-sr-sim can run as:

• a standalone Python CLI for local sizing and what-if analysis
• an HTTP service with vllm-sr-sim serve
• a sidecar container that vllm-sr serve starts by default on the shared vllm-sr-network

Read this section in order

1. Getting started for local sidecar, standalone CLI, and external service setup
2. Dashboard integration for the proxy path and UI surfaces
3. Capacity planning scenarios for example-driven decision workflows
4. Simulation model reference and power model reference when you need the underlying mechanics
5. Guide PDF and guide assets when you want the printable version or source files