Login
Computer Science @ Rochester
Thursday, May 02, 2013
1:00 PM
CSB 703
Ph.D. Thesis Proposal
Zhuan Chen
Computer Science
Virtualization for Data-Driven Applications on Low-Power Systems
Virtualization for Data-Driven Applications on Low-Power Systems

Technology evolution today brings advanced hardware to low-power systems, such as many development boards, smartphones, and tablets, and enables the ability to support more sophisticated software on these platforms. Low-power systems, with certain high computational capacity and ever low power consumption, are becoming another potential computing platform and important complement to traditional full-power machines in certain areas. The thesis of this proposal is to study the usage scenarios enabled by those low-power systems and explore the corresponding new system infrastructure supports.

The computational capability and platform mobility exhibited by today's low-power systems bring the potential of self-sustainable computing in the field deployment. Data collection and information learning in remote fields are increasingly important for applications in highway traffic management, area surveillance, and environmental sciences. The introduction of advanced computing to the remote on-scene scenario will substantially enhance the capability of data-driven services in the field and enable more wide-range application of field systems. At the same time, this future trend also brings new challenges, like the remote application deployment and data security concern in the field. We advocate the leverage of virtualization, which is widely used in today's data centers and cloud computing, to address those challenges and build future field system infrastructures. While virtualization can serve as the promising technique support, it also needs to tackle the conflict arising between the extra overhead induced from virtualization and the constraints of limited resources on low-power platforms. We further propose reservation mechanisms that allow specialized VMs to attain near-native performance at particular resources including CPU time, device accesses, and memory management.

We also look at another usage scenario of low-power systems for data-center workloads. Academies and industries today have both started to take advantage of low-power platforms in data centers for improved energy efficiency compared with conventional server machines. We are further exploring new techniques for better support of virtualization on low-power systems to make them truly ready to use in the cloud service platform.