Project: Virtual Machine Disk I/O Reduction via Hypervisor-Level Prefetching.
Disk I/O prefetching has been commonly used in current operating systems to improve the I/O efficiency of guest applications. In this project, we use hypervisor-level prefetching as a complement for guest OS prefetching to achieve further I/O throughput improvement. We examine the effectiveness of hypervisor-level prefetching under different guest operating systems and various disk I/O-intensive benchmarks.
Project: Printing solution recommendation based on highly incomplete datasets (US patent pending approval).
This system provides new users with recommended printing workflow solutions based on accumulated system usage history. Unlike other recommendation systems, we have to deal with highly sparse data since customers are typically lazy or incapable of giving complete and accurate answers to printing facility specific questions. We use machine learning techniques to project missing data, find hidden customer patterns, and generate solution recommendations.
Project: A Java-based simulator of high-speed particle accelerators.
Based on fast numerical algorithms (e.g., Fast Fourier Transform, the solver of linear equations and partial derivative equations), the simulator approximates the behavior of particle beams in accelerators.