Toward to Utilize the Heterogeneous Multiple Processors of the Chip Multiprocessor Architecture

摘要

Continuous improvements in semiconductor fabrication density are supporting new classes of Chip Multiprocessor (CMP) architectures that combine extensive processing logic/processor with high-density memory in a single chip. One of the architecture, called Processor-in-Memory (PIM) can support high-performance computing by combining various processors in a single system. Therefore, a new strategy is developed to identify their capabilities and dispatch the most appropriate jobs to them in order to exploit them fully. This paper presents a novel scheduling mechanism, called Swing Scheduling to fully utilize all of the heterogeneous processors in the PIM architecture. Integrated with our Octans system, this mechanism can decompose the original program into blocks and can produce a feasible execution schedule for the host and memory processors, even for other CMP architectures. The experimental results for real benchmarks are also proposed.