Saturation NOW

We present a distributed version of the saturation algorithm for symbolic state-space generation of discrete-state models. The execution is strictly sequential but utilizes the overall available memory. A level-based allocation of the decision diagram nodes onto the workstations is created. No additional node or work is created. A dynamic memory load balancing heuristic helps coping with the uneven growth of the decision diagram levels allocated to each workstation. Experiments on a conventional network of workstations show that the runtime of our distributed implementation is close to the sequential one even when balancing is triggered, while it is of course much better when the sequential implementation is forced to rely on virtual memory.