Towards Parallel Model Generation for Random Performance Testing of Model-Oriented Operations

Model-oriented operations, such as model transformation, model query, and model comparison, are the core of Model-Driven Engineering. Their scalability becomes an important issue when they are to be applied in industry. Large-scale models, however, are not widely available, making it hard to test the performance and coverage of those operations without any bias. To do so, one must be able to generate large models, which are syntactically correct, as test inputs efficiently and randomly. This paper proposes a parallel approach to generating large random model, which improves on our previous sequential algorithm. First, the paper identifies the dependencies existing in the process of model generation, which hinders parallelization. Then, the paper proposes a partitioning strategy that is able to turn a sequential generation task into a series of parallelizable subtasks. Experimental results of performance are also presented, which show that the parallel approach proposed can reduce 50% time costs compared to the sequential one in most cases, while the correctness and the randomness of the models generated by this approach are kept.