Tailored Preloading, Precaching and Prefetching Loading Strategies for Applications Through a Multi-Component AI

Sammanfattning: When using a personal computer the user could experience starting an application for the first time, loading a level in a video game or any tedious arbitrary loading process. These loading processes often load mostly the same data each time and consume time where the user is waiting. Making it faster is often preferred. The file-caching tactics Preloading, Precaching and Prefetching (PPP) have received past research, mostly to make generic improvements or better algorithms. The gap in the work is a lack of non-generic file-caching optimisation algorithms. To improve loading times during a Mostly Deterministic File-Loading Process (MDFLP), this research suggests a Multi-Component Artificial Intelligence (MCAI) that analyses the application's runtime results. The proposed MCAI would not be a one size fits all kind of solution but rather generate a tailored File-Loading Strategy (FLS) for the application. The aim of the research is to investigate if a MCAI could improve loading speeds of arbitrary application's MDFLPs. The objectives are to implement a test synthesiser for generating synthetic test environments to use MCAI on, to implement the MCAI and perform the experiments. The research questions regard how the MCAI can analyse inefficient operations during a MDFLP, propose measures that increase efficiency and aid developers independently of the high-level technology used. They also regard how the MCAI through iterative runs of an application can generate an application-specific FLS which is better in terms of PPP performance. The method section goes into detail about how the test synthesiser, the MCAI and the host application is implemented. It also explains how the experiment was made, what the experiment tested, what data was collected and what hardware and software were used. The result first shows in detail how the MCAI works on a simple test and then moves on to three extensive tests. Two of the tests show positive results, where the MCAI manages to generate an optimal FLS, whilst the MCAI fails in the third test. The third test highlights inherent weaknesses in the MCAI. The conclusion is that the MCAI shows potential. In its weakest form, it still manages to produce good results, where the MCAI generates a FLS that improves the load time performance. There is improvement potential for the MCAI to make it more smart, efficient, reliable and make it able to generate a FLS for test three. The research leaves room for follows ups and projects, like developing the MCAI further and performing case studies.