(A) memory managing scheme using memory access patterns of multimedia

Abstract

Low memory embedded systems, such as mobile devices or PMPs, are now very widely used in our lives. However, in multitasking environments, programs often need to wait for pages to be loaded on the main memory, especially for those which were on the main memory once but later swapped out due to memory shortage. This is what actually happens when a page fault occurs, and often the main hindrance to program’s seamless execution. We implemented Page Swap Monitor, which enables quantitative analysis of paging behaviors. Using the tool guided us that page faults in heap are the main reason of paging delays. We focused on multimedia applications, especially movie players. Our aim is to provide seamless execution of multimedia players in low-memory multitasking environments. In this paper, we propose a scheme that helps multimedia players perform better and execute with consistent quality over other programs. In this paper we introduce the notion of stream, memory allocations with the same function call backtrace and memory accesses to them. We discovered memory access pattern information in each stream from profiled access traces, and used a new memory allocation scheme and prefetching/pinning scheme during runtime based on the discovered pattern information. Our method helped reduce the number of page faults in heap, and provided better execution quality for multimedia players.