GDCSim

Consuming between 2-3% of the U.S. national power grid, data centers are under increasingly greater pressure to cut energy costs and reduce their carbon emissions. Operational infrastructure, such as power management and cooling, can account for over half of a facilitys energy usage. Current methods to mitigate infrastructure costs include energy-efficient resource management, such as algorithms for workload scheduling and power allocation, and improving a facilitys physical design by thermal modeling using computational fluid dynamics (CFD). CFD simulations are computationally intensive and can take hours to complete, making them cumbersome for processing any time-dependent response. Existing online resource management algorithms ignore physical conditions, creating hotspots that prompt a redesign of the data centers layout by CFD simulations. Thus, an approach is needed that considers such cyber-physical interdependencies. Researchers at ASU have developed GDCSim, a simulator that models online resource management with environmental behavior and accurately predicts performance as well as energy consumption of a data center. There are three components to GDCSim: a dynamic CFD emulator, multi-tier resource manager, and a simulator that provides immediate feedback based on different resource management designs. It emulates CFD simulations to within 92% accuracy in a fraction of the time and seamlessly integrates with any workload scheduling software or power management platform. The simulator can be used to design a data centers layout based on thermal imaging, to develop physically-aware resource management algorithms, and conduct performance analyses of different infrastructural configurations. The different components of GDCSim can be used independently, or all together for a robust cyber-physical representation. Potential Applications Data Center Infrastructure Management (DCIM) Facility Performance Analysis Workload Scheduling Thermal Modeling Benefits and Advantages Comprehensive Accounts for both electronic and physical aspects of DCIM and their interdependencies. Efficient Cuts modeling and simulation time without sacrificing accuracy. Lower Costs/Carbon Emissions Enables users to create a more optimal design, saving money by reducing electricity use. Retrofit Integrates with existing resource management software. Versatile GDCSims multiple components can perform all facets of DCIM autonomously as well as holistically. Download Original PDF For more information about the inventor(s) and their research, please see Dr. Sandeep Gupta's directory webpage Dr. Georgios Varsamopoulos's directory webpage