SANgo

SANgo simulates storage area networks using discrete-event modeling to analyze the behavior and stability of complex storage infrastructures and is implemented in the Go programming language.

Key Features:

• Discrete-Event Modeling Paradigm: Employs discrete-event modeling to capture structural and dynamic aspects of high-level storage system components.
• Configurable Architecture: Supports customizable numbers of components to replicate diverse storage configurations.
• Granular Simulation Control: Allows definition of simulation granularity based on replicated patterns observed in real systems to explore stability boundaries.
• Monitoring Interfaces: Provides interfaces for monitoring metrics of storage controllers, network connections, and hard drives.
• Real-Time Parameter Tuning: Enables dynamic adjustment of simulation parameter values during runtime to support realistic digital twins.
• Reinforcement Learning Integration: Supports reinforcement learning approaches to train models that minimize discrepancies between simulated and actual SAN data via external parameter adjustments.
• OpenAI Gym Interface Support: Offers compatibility with the OpenAI Gym interface for benchmarking learning algorithms.

Scientific Applications:

• Stability Analysis of Storage Systems: Investigates stability limits and potential failure points of real-world storage systems through simulation.
• Digital Twin Calibration: Calibrates simulator parameters and creates digital twins by tuning parameters in real time to align simulated outputs with observed SAN data.
• Machine Learning Benchmarking and RL Training: Trains and benchmarks reinforcement learning and other learning algorithms to minimize simulator-to-real data discrepancies using the OpenAI Gym interface.

Methodology:

Implements discrete-event modeling and exposes monitoring interfaces for storage controllers, network connections, and hard drives; supports real-time parameter tuning, reinforcement learning–based parameter optimization to align simulation with SAN data, and an OpenAI Gym–compatible interface for benchmarking.