Effectiveness of Random Neighborhood Selection at Scale

Investigate whether selecting random neighborhoods for OrigaMIG's restricted binary integer linear programming calls performs as effectively as Algorithm 1 at higher offered loads and in larger data centers.

Background

OrigaMIG selects a bounded neighborhood of physical machines and solves a restricted binary integer linear program over that neighborhood. In the evaluation at an offered load of 0.85, randomly selected neighborhoods achieved nearly the same number of active physical machines as the neighborhoods chosen by Algorithm 1, suggesting that the computationally cheaper random selection strategy might be sufficient in that setting.

The paper does not establish whether this result generalizes to more heavily loaded systems or larger data centers. Determining the robustness of random neighborhood selection under those conditions would clarify whether OrigaMIG can simplify its neighborhood-selection mechanism without sacrificing placement quality.