Cause of diminishing parallel speedup

Determine whether the diminishing returns in the parallel exact algorithm for $(\ell,d)$-Correlated Motif Mining are caused by load imbalance among wildcard-combination jobs, with an expensive combination dominating the wall-clock time at high thread counts.

Background

The exact algorithm parallelizes independently over pairs of wildcard-position combinations. Experiments show near-linear speedup through 32 threads, but reduced gains beyond that point: the speedup peaks at 64 threads and does not improve further at 128 threads.

The authors explicitly state a conjectural explanation rather than establishing the cause experimentally. Resolving this issue would clarify whether load balancing or another implementation or systems-level factor limits the scalability of the parallel CMM algorithm.

References

We conjecture that these diminishing returns are due to load imbalance: some wildcard combinations are more costly than others, and at 128 threads, an expensive combination dominates the wall-clock time.

— Mining Correlated Motifs with Wildcards in PPI Networks: Exact and Optimal  (2610.17958 - Oliveira et al., 13 Oct 2026) in Section 5.5, Results for RQ5