An Improved Approach to Tag Reduction on Low Power CMP with Trade-Off of Energy and Performance

Energy consumption and performance are always major considerations in microprocessor optimization. However, mostly, saving energy causes performance decrement. Based on the previous energy saving research on tag reduction for low power Chip Multiprocessor (CMP), we proposed an improved approach to balance energy and performance by introducing an affiliation degree. To our best knowledge, this is the first work that analyzes the energy and performance of tag reduction on CMP. We also formulate the trade-off of energy and performance to an equivalent problem which is to find out an appropriate affiliation degree to get the benefit with trade-off of energy and performance most. In the experiment, we apply our approach to a 16-core CMP with help of simulation. The experiment results show that when affiliation degree is 12, the energy saving and performance cost of tag reduction on CMP can be balanced best.