CISE Postdoctoral Research Associates in Experimental Computer Science: Experimental Evaluation of Active Caches For Fine Grain Multi-Threading Systems

点击翻译按钮获取中文摘要

9901553Gao, Guang R.University of DelawareCISE Postdoctoral Associates in Experimental Computer Science: ExperimentalEvaluation of Active Caches for Fine Grain Multi-Threading SystemsThis research will investigate the effect of active cache memories on computer system performance. An active cache is capable of controlling complex data transformations and data movements under the direction of a programmer or compiler. This study, performed through system simulation, requires designing active caches, proposing performance models for the caches, and developing compiler mechanisms to support the percolation model of execution.A percolation model of execution has been proposed in which the application requests that some data transformations be performed and that the transformed data be transferred to a higher level in the memory hierarchy. A thread becomes ready for execution when the data that it needs has percolated. The active caching mechanism is necessary in systems in which the latency to fetch a piece of data in a lower level of the memory hierarchy is much higher than the time required to process the data.The effectiveness of caches that actively percolate data closer to the processing unit before the execution of a piece of code starts is being proposed. This model is different from traditional pre-fetching because itis not based in any prediction mechanism: the instructions requesting the percolation of a data block are inserted into the code either under the programmer direction or through compiler technology. It is expecedt that the experimental study that demonstrates the viability of this idea will spur further research in the area of active hardware system components.