Over the past decade, the confluence of an unprecedented growth in data volumes and the rapid rise of cloud computing has fundamentally transformed systems software and corresponding infrastructure. To deal with massive datasets, more and more applications today are scaling out to large datacenters. These distributed data-parallel applications run on tens to thousands of machines in parallel to exploit I/O parallelism, and they enable a wide variety of use cases, including interactive analysis, SQL queries, machine learning, and graph processing. Communication between the distributed computation tasks of these applications often result in massive data transfers over the network. Consequently, concentrated efforts in both industry and academia have gone into building high-capacity, low-latency datacenter networks at scale. At the same time, researchers and practitioners have proposed a wide variety of solutions to minimize flow completion times or to ensure per-flow fairness based on the point-to-point flow abstraction that forms the basis of the TCP/IP stack. We observe that despite rapid innovations in both applications and infrastructure, application- and network-level goals are moving further apart. Data-parallel applications care about all their flows, but today’s networks treat each point-to-point flow independently. This fundamental mismatch has resulted in complex point solutions for application developers, a myriad of configuration options for end users, and an overall loss of performance. The key contribution of this dissertation is bridging this gap between application-level performance and network-level optimizations through the coflow abstraction. Each multipoint-to-multipoint coflow represents a collection of flows with a common application-level performance objective, enabling application-aware decision making in the network. We describe complete solutions including architectures, algorithms, and implementations that apply coflows to multiple scenarios using central coordination, and we demonstrate through large-scale cloud deployments and trace-driven simulations that simply knowing how flows relate to each other is enough for better network scheduling, meeting more deadlines, and providing higher performance isolation than what is otherwise possible using today’s application-agnostic solutions. In addition to performance improvements, coflows allow us to consolidate communication optimizations across multiple applications, simplifying software development and relieving end users from parameter tuning. On the theoretical front, we discover and characterize for the first time the concurrent open shop scheduling with coupled resources family of problems. Because any flow is also a coflow with just one flow, coflows and coflow-based solutions presented in this dissertation generalize a large body of work in both networking and scheduling literatures.