Data Marshaling for Multicore Systems
Muhammad Suleman, Onur Mutlu, Jose Joao, Khubaib Khubaib, Yale N. Patt · IEEE Micro · 2010
Dividing a program into segments and executing each segment at the core best suited to run it can improve performance and save power. When consecutive segments run on different cores, accesses to intersegment data incur cache misses. Data Marshaling eliminates such cache misses by identifying and marshaling the necessary intersegment data when a segment is shipped to a remote core.