3 ms·

For many types of workloads (like AI inference), high availability is not needed for individual racks.