Designing Next‐Generation Computing
We are witnessing a great wave of digitized data, with widespread use of smart phones, increasing scientific and medical data generated in digital format, and the variety of information collected in cyber-physical systems. Typically the data are generated more dynamically and at a much higher rate than previously, and they demand more intelligent interpretation and retrieval mechanisms.
The computing system that has arisen in response is the cloud. This legacy architecture was based originally on proprietary search engines, e-commerce systems, and grid computing. It has since been upgraded with ad hoc changes, but fundamental problems remain in the areas of scalability, response quality and speed, and energy consumption.
During her Center appointment Professor Lu will work toward designing a computing system that is amenable to big data, is scalable and energy-efficient, and has performance guarantees.
The project consists of four parts:
1. Design novel task-scheduling architecture and algorithms to improve response time by orders of magnitude without sacrificing quality or consuming extra resources. An equivalent interpretation is achieving the same response time and quality with only a fraction of the energy consumption.
2. Design data-placement architecture and algorithms to improve interpretation and retrieval, and to achieve scalability of the system.
3. Measure and analyze workload to understand characteristics in production clusters, with efficient tracing to allow fast and reliable evaluation of a large system in a small testbed.
4. Balance loads to reduce the power-conversion loss (currently as high as 10-15 percent before computation even takes place) using a series-stacked power-delivery architecture.
In pursuing her project goals, Professor Lu is collaborating with colleagues in power electronics, circuits, medical imaging, and genomics.