IP Deadline-Aware Bank Partition Scheme for Memory Scheduler in Mobile Platform

초록

In a mobile platform, diverse IP cores are integrated and share memory bandwidth. As frame-based applications (audio, video, and camera) use large volume of data, they cause severe interference while accessing memory. Severe interferences cause IP cores to miss their deadlines and slow down applications running on CPU cores. In this paper, we propose IP deadline-aware bank partitioning to simultaneously guarantee IP's deadline and enhance CPU's system performance. First, we prioritize CPU memory requests to improve system performance. Second, we allocate the proper number of banks to IPs so that the estimated bandwidth of allocated banks is higher than IP's maximum bandwidth requirement. Experimental results show that the proposed scheme significantly improves CPU performance with guaranteeing IP's deadline.

키워드