Combining Explicit Flow Measurements and Network Programmability for Network Management

The Internet is indispensable to modern society, yet its management remains complex. To facilitate common solutions, its constituent networks have gradually converged on a set of core protocols, implicitly reinforcing the status quo and limiting innovation. The emergence of QUIC challenges this trajectory and revitalizes the Internet. However, it also disrupts a well-established ecosystem and complicates network management. In particular, operators face two key challenges: first, QUIC exposes limited information, necessitating new methods to assess network performance; second, QUIC's user-space nature leads to more diversity in protocol behavior, requiring more adaptive and flexible networks. This dissertation shows, through a combination of theoretical analysis, simulations, testbed experiments, and large-scale Internet measurements, that two complementary technologies - explicit flow measurements (EFM) and network programmability - can efficiently address these challenges. On the one hand, we demonstrate that EFM can accurately capture QUIC performance, that the latency spin bit is already used by real-world web traffic, and that EFM enables systematic network assessment through tomography. On the other hand, we show that network programmability accelerates developing novel solutions and enhances the flexibility of modern networks. Concretely, we investigate how EFM can be monitored on programmable hardware, identify challenges in implementing congestion management on such platforms, and pioneer congestion management algorithms that leverage EFM to identify misbehaving flows and shield benign ones. Overall, this work establishes EFM and network programmability as key enablers for effective network management in a revitalized and increasingly complex Internet.