“…Although reverse-search based algorithms can guarantee that the time cost between two consecutive outputs is bounded by a polynomial function of the input size (i.e., polynomial delay), which implies that the time complexity of these algorithm is proportional to the number of maximal cliques within the input graph (i.e., output-sensitive), their performances on real-world graphs are not as efficient as those of branch-and-bound based algorithms [44]. For the latter, there are also numerous studies that develop parallel algorithms for MCE, in which [64,71] target the data-level parallelization with SIMD instructions, [5,22,36,45,47,91,104,135,171,173] target the task-level parallelization with shared-memory setting and [24,32,69,148] target the tasklevel parallelization with distributed-memory setting. Basically, these parallel algorithms extend the techniques that are originally proposed for sequential MCE algorithms and aim to balance and reduce the overheads in task assignment and memory management to achieve higher degrees of parallelism.…”