NI-ORCA Parallel Algorithm Counts Non-Induced Graphlet Orbits 30x Faster
A new parallel algorithm achieves up to 30x speedup over sequential methods for exact non-induced orbit counting.
Counting graphlet orbits is critical for understanding vertex roles in network analysis, but existing algorithms only efficiently handle induced graphlets. Many real-world applications—such as bioinformatics, social network analysis, and cybersecurity—require exact, scalable non-induced orbit counts, for which no parallel method existed. NI-ORCA fills this gap by extending the ORCA framework with a reformulated system of linear equations, enabling exact counting of non-induced graphlets up to K4 (4-clique). The algorithm operates in three stages: triangle counting, 4-clique enumeration, and orbit solving, each with custom parallelization strategies using thread-local memory and data structures to reduce contention and balance workload.
Experimental results on eight real-world graphs and a series of synthetic Erdős–Rényi networks show that NI-ORCA significantly outperforms existing sequential algorithms, delivering up to 30x speedup. The authors also analyzed scheduling policies, chunk sizes, and affinity strategies, finding that dynamic scheduling with small chunk sizes provides the best performance and load balancing. NI-ORCA represents a major step forward for scalable, exact non-induced orbit counting, enabling deeper insights into the structural roles of vertices in large graphs.
- NI-ORCA is the first parallel algorithm for exact counting of non-induced graphlet orbits up to 4-clique.
- Achieves up to 30x speedup over state-of-the-art sequential algorithms on real-world datasets.
- Uses stage-specific parallelization with thread-local memory to minimize contention and balance workloads.
Why It Matters
Enables scalable, exact analysis of vertex structural roles in large graphs—critical for bioinformatics, social networks, and cybersecurity.