Abstract
We present an alternative to traditional approaches of parallel and distributed model verification. In contrast to methods based on partitioning the state space, we keep the states together, sorted and organized as a stream.
The stream is passed across computational nodes in a logical circle where each node updates the states in the part of the stream it currently sees. The presented method relies on state locality, allows storing the state space at external storage devices, and scales well on a non-dedicated, non-uniform cluster of up to tens of nodes.
We have implemented an experimental version of the algorithm and obtained very promising results in scalability in the number of processors and computational nodes used in a large real-life use case.