Research on the speculative Processing of computer systems based on the information theory approach

点击翻译按钮获取中文摘要

This research examined the various universal data prediction techniques argued in the field of information theory as application to the speculative processing in computer systems. Two research groups, the information theory group and the computer architecture group, jointly carried out the research and build the data structure suitable for the speculative processing in computer systems. Moreover, we designed the speculation execution architecture which uses the proposed data structure, and evaluated its performance. The main results of this research are as follows.1. Antidictionary data structure suitable for speculative processingThe compressing method called the anti-dictionary using the set of the minimum series which does not appear for a data series was examined. We present a fast and memory-efficient algorithm to construct an antidictionary for a binary string using a suffix tree. It is proved that the complexity of the algorithm is linear in space and time, and its effectiveness is demonstrated by simulation results.2. Branch prediction for microprocessorsThe branch prediction scheme using the pattern matching with a branch history and an instruction address was proposed, and its prediction accuracy was examined. We verified the good prediction accuracy of the scheme compared with the conventional branch prediction only using a branch history when there was no restriction of computational complexity or the amount of hardw are resources.3. Switching router with predictionsIn order to realize the predictive switching router which operates efficiently, we proposed the technique of reducing prediction mistake packets, and the technology of detecting and canceling a prediction mistake packet. We performed the evaluation by a simulation about the effect of predictive switching in consideration of a dynamic communication collision and verified the high performance of the proposed technique.