A classification-based prefetching scheme for hibernation of embedded computing systems

With the rapid growth of embedded computing system markets, e.g., intelligent home appliances and smart TVs, vendors and researchers are developing more user-friendly interfaces and seeking to provide more sophisticated applications with better functionalities. Such a developing trend would prolong the initialization time of these embedded computing systems. Hibernation (or suspend-to-disk) that retains a computing system's state after power recycling is regarded as a solution to reduce the booting time of systems and applications to meet the requirement of user experiences. In contrast to the existing hibernation techniques that dump most of the memory pages to the secondary storage, we propose a classification-based prefetching scheme to improve the system performance on both of the hibernation and resuming with minimized I/O overheads by jointly considering the system/application behaviors and the usage patterns of memory pages. The proposed scheme was also implemented with the Linux kernel on an evaluation board to show the capability of the proposed scheme.