A Classification Model for Predicting Suitable Cache Level in a Multi-core Architecture

Cache memory has an important role in achieving system performance in multi-core architecture. Finding the best suitable cache configuration for an application is a very important step while designing a computer system to improve the performance. The commonly considered cache memory design parameters are the size of the cache, line size, associativity, type of cache, replacement policy, writes policy and levels of cache. Selecting these parameters decides the design goals such as system performance, energy consumption, area, scalability and programmability of an application. Cache design space is a time consuming and complex process as it involves studies on the impact of all possible cache parameter configurations on system performance. This project mainly aims at creating a model that can predict an efficient cache configuration -cache level- that is best suitable for an application in terms of energy consumption and performance in a multi-core environment using machine learning techniques such as classification. The design goal here is to predict the optimum cache levels for an application by considering the design parameters such as cache sizes, associativity, block size etc. This method will be carried out in a multi-core environment for the studies on advancements in computer architecture. The required data set is generated using two simulators such as Gem5 and CACTI. The entire process of data collection is automated via shell scripting in Unix OS and applications from different domains will be considered here with different cache parameter combinations. The performance of the classifier is measured based on the evaluation metrics such as Precision, Recall, and F- measure. Performance measurement concerning power consumption and execution time would be the figure of merits of this project.