Scheduling page migrations using latency tolerance

Abstract

Solutions for scheduling page migrations use latency tolerance of coupled devices, such as external peripheral devices (e.g., network adapters), to prevent buffer overflows or other negative performance. A latency tolerance of a device coupled to a virtual object, such as a virtual machine (VM) is determined. This may include the device exposing its latency tolerance using latency tolerance reporting (LTR). When a page migration for the virtual object is pending, a determination is made whether sufficient time exists to perform the page migration, based on at least the latency tolerance of the device. The page migration is performed if sufficient time exists. Otherwise, the page migration is delayed. In some examples, latency tolerances of multiple devices are considered. In some examples, multiple page migrations are performed contemporaneously, based on latency tolerances. Various options are disclosed, such as the page migration being performed by the virtual object software or the device.