Prediction and Grouping
Most systems treat each data request as an independent event. However, such requests in a computer system are driven by programs and user behaviour, and are therefore far from random. We have conducted research into multiple aspects of predicting data access behavior, and the identification and exploitation of such behavior to produce: predictive caches, improved access predictors, informed data layout, and the automated grouping of related data. This work is ongoing, and has also been extended to problems in mobile data management, file migration, data de-duplication, and power conservation.
Despite having resulted in new avenues of research on mobile storage management, work is ongoing on file and data access prediction. We have developed new access predictors that incorporate machine learning technique to automatically improve prediction quality and reduce the chances of mispredictions. We have also used machine learning to adaptively place files in non-hierarchical storage systems.
We are currently using statistical and machine learning techniques to group interleaved, metadata poor trace data and investigating how different groupings apply to application areas including fault isolation, power management, and de-duplication.