Overall Architecture

The overall architecture of CSLE is shown in the Fig. 2

Figure 2: Architecture of CSLE; it consists of an emulation system (bottom part), a management system (middle part), and a simulation system (upper part); the three systems are connected through APIs and a shared metastore.

In the middle of Fig. 2 is the management system, which orchestrates the overall execution of the framework and stores metadata in a distributed database called the metastore. It has three interfaces. One interface is a gRPC API to the emulation system, which emulates IT infrastructures using hardware and network virtualization. The other two interfaces—a web interface and a CLI—are to the simulation system, which simulates Markov decision processes and runs reinforcement learning workloads.

Users access CSLE in three ways: (1) through software functions using Python libraries of the simulation system; (2) through a web browser using the web interface; and (3) through a terminal using the CLI.