The rapid growth of computing in data centers worldwide has led to a significant increase in energy consumption and carbon emissions. As the demand for cloud services continues to surge, data centers have become the backbone of the digital economy, supporting a wide range of applications from social media and e-commerce to scientific research and artificial intelligence. This exponential growth in data centers has driven the need for more servers, storage devices, and networking equipment, all of which consume substantial amounts of electricity. This in turn contributes to higher carbon emissions and environmental impact. To make things worse, the end of Moore’s law and Dennard scaling has exacerbated scaling challenges in both computing units like CPUs and network infrastructure, creating a bottleneck for data center growth and efficiency.

In this thesis, I propose two new ways to address scalability challenges in data centers and to support future data center growth:First, I propose P-Net, a novel datacenter network architecture that improves the efficiency and scalability of data center networks (DCNs). P-Net, or parallel dataplane network, employs multiple network planes to achieve higher bandwidth and lower latency, instead of relying on the free scaling of underlying chips in traditional networks. By explicitly leveraging network chip parallelism, P-Net enables linear scaling of network resources with respect to bandwidth, reducing energy consumption, monetary costs, and scalability challenges in DCNs. I then focus on the growing energy and carbon footprint of computing units like CPUs and accelerators, and explore new ways to tackle these challenges by exploiting low-carbon renewable energy sources. In particular, I explore the feasibility of space-shifting computational workloads across data centers as a carbon-aware computing solution. By utilizing energy efficient fiber optics wide-area network (WAN) links, space-shifting can make better use of localized renewable energy sources that cannot be used otherwise, achieving the goal of lowering power demand on the grid and reducing carbon footprint. Through detailed analysis, I show that by selecting the optimal data center(s) to run workloads based on both local carbon intensity and WAN cost, space-shifting can be more effective than alternative time-shifting solutions.