A hyper-converged infrastructure (HCI) provides one or more HCI units, each HCI unit including a pair of physical computing servers coupled to shared, nonvolatile storage. The shared, nonvolatile storage of each HCI unit is dedicated to that pair of physical computing servers and provides redundant storage of application data. Each of the pair of physical computing servers runs a set of application instances and an IO (Input/Output) stack. The application instances receive client requests over a network and generate IO requests specifying reads and writes of application data. The IO stack processes the IO requests to effect the specified reads and writes in the shared non-volatile storage.