This paper proposes an algorithm for disaster avoidance control against heavy rainfall. According to weather information, the algorithm reconfigures a logical network (slice), including the migration of virtual machines (VMs), to avoid disasters. It was applied to a nationwide network of 105 nodes and 140 edges, including cases with more than 10 000 slices. Through numerical simulations using actual data of rainfall that caused significant damage in Japan, we found that the probability of service disruption under the proposed control with suitable parameter settings is 10%–30% of that without control, on average. Our proposed control experimental system is implemented by using the software-defined network technology. It can migrate VMs and estimates VM migration time to determine how many VMs should be migrated. By using the experimental system, we found that the control interval has an optimal value, which depends on the management system processing capacity.