I have been working with Professor Nikil Dutt at UC Irvine since 2013. My research focuses on managing resources at runtime for diverse workloads executing on many-core and heteregeneous systems. My work balances multiple objectives (e.g., performance, energy efficiency) by monitoring workload behavior and operating conditions at runtime, and configuring the system appropriately. Decisions are made by utilizing adaptive policies and self-awareness. During my thesis work I focused on the memory subsystem and heterogeneous multiprocessors, but have expanded to include SoCs and IoT networks. I am interested in applying my resource management techniques to networks of smart devices and beyond.
Projects I am currently involved in are listed below. For more detail, you can see my research vision.
MARS consists of a toolchain for creating resource managers that allows users to easily compose models and policies that interact in a hierarchy defined by the granularity of the actuations performed in the system. MARS is implemented and evaluated on top of a real Linux-based platform. Furthermore, MARS also provides an offline simulation infrastructure for fast prototyping of policies and large-scale or long-term policy evaluation.
My contributions to MARS include extensions to support a custom offline simulator, as well as the gem5 architectural simulator. I have used MARS to deploy and evaluate resource management policies for both simulated multicores and real Linux platforms such as the ODROID XU3.
The project exploits self-awareness principles, together with lessons learned from large-scale factories to contain complexity, achieve predictability and manage robust system design. The overall research theme will demonstrate the utility of self-aware IPFs in managing MPSoC complexity, while achieving scalability, predictability, and system efficiency, with the long term goal of supporting autonomous systems as a main application.
My work in this area focuses on providing adaptive resource management using hierarchical control and goal-driven autonomy.