For a given number of cities and approximately equal clusters, the larger the number of clusters P, the faster the IGA (Section 2.2). On the other hand, larger P results in a more local view of optimization with a large number of (potentially critical) tours not being explored. From our experiments, genetic algorithms do well on small TSPs with up to 100 cities. Therefore, when decomposing, it is appropriate to keep each cluster within 60 - 100 cities.