Zhu, Crainic, and Gendreau (2011) compare the performance of
their hybrid algorithm with a state-of-the-art solver for small to
medium sizes of a randomly generated data set. They show that
their algorithm outperforms in computational time and even solution
quality when instance size grows. For instance, the combination
of slope scaling and long-term memory-based perturbation
achieves on average 21% improvement in solution gap in 10 hours
compared to the solver. Inclusion of ellipsoidal search even im-
proves the solution gap by 2% more. For small instance, this hybrid
algorithm finds the optimal solution of 5 out of 7 instances and
reaches a gap of 0.13% for the others.