The goal of building energy-efficient information and communication networks is now widely embraced. Various mechanisms are being developed to save energy at network nodes based on local traffic conditions and objectives; however, their cumulative impact on end-to-end application performance is not well understood. TCP is the dominant transport protocol on the internet, and its end-to-end operation ties in the performance of all energy saving mechanisms in the network. We develop a network utility theory to study and design distributed algorithms for network congestion control and energy consumption optimization, and study their stability and convergence properties. For traditional TCP algorithms and active queue management, we propose a novel dynamic bandwidth adjustment mechanism to turn off excess link resources and save energy without compromising end-to-end performance. For the emerging multipath TCP, we propose new energy-aware source algorithms which maximize throughput with applications to multi-radio smartphones and datacenter networking.