Traditional solutions for high-performance computing batch schedule computation on first available resources, leading to hot-spots and inefficient heat flows within the datacenter infrastructure, overworked equipment with costly failures, and a maintenance nightmare. Hence, the first step for bringing energy efficiency into a datacenter is to be able to see in real-time how it actually works, in minute detail.GLIMPSE
* is our decision support system for the GreenLight datacenter. It carefully correlates data from thousands of sensors and assembles domain-specific views tailored for a diverse set of stakeholders. This approach reduces personnel training costs, eases adoption, and minimizes the risks of operator failures.
It also serves as a centerpiece for the second step: helping our researchers profile applications and analyze power and environmental data so they can develop innovative heat/energy-aware spatiotemporal scheduling. This scheduling strategy incorporates an understanding of the topology and real-time heat distribution of the computation, while also balancing equipment load to reduce failure, optimize power usage, minimize necessary cooling, and improve overall performance.
Overall, GLIMPSE showcases the potential of the GreenLight project, brings energy awareness to all interested parties, and helps promote UCSD's leadership position in "green" technologies.
* GLIMPSE is optimized for touch displays and tablets and as such has "sliding" horizontal imagery. To reach this material on none-touch devices use the horizontal scroll bars usually found at the bottom of the browser. Supported browsers include current versions of Firefox, Safari and Internet Explorer.