This paper addresses the critical issue of user-perceived quality of experience (QoE) in Internet video applications, which significantly impacts content provider revenues. Given the lack of support for optimizing QoE measures in the network, a robust bitrate adaptation algorithm in client-side players is essential to ensure a good user experience. The paper makes three key contributions: (1) developing a principled control-theoretic model to reason about a broad spectrum of strategies, (2) proposing a novel model predictive control (MPC) algorithm that optimally combines throughput and buffer occupancy information, and (3) presenting a practical implementation in a reference video player, dash.js, to validate the approach using realistic trace-driven emulations. The authors first formulate the video bitrate adaptation problem as a stochastic optimal control problem, defining key dynamic variables and an objective function. They identify the limitations of existing approaches that rely solely on rate- or buffer-based strategies and argue that combining both signals can lead to better performance. The MPC approach predicts the expected throughput for the next few chunks and uses this information to make optimal bitrate decisions for QoE maximization. To address the computational overhead and practical challenges, they develop a fast and low-overhead implementation, FastMPC, which uses a table enumeration approach to store optimal control decisions for future online use. The evaluation section compares the proposed approach against existing rate- and buffer-based algorithms using real and synthetic throughput variability traces. The results show that the proposed MPC approach consistently outperforms state-of-the-art adaptation algorithms by 15% in broadband datasets and 10% in cellular datasets in terms of median QoE. The FastMPC implementation adds negligible overhead to the baseline dash.js player, with only 60 kB extra memory usage and similar CPU usage.This paper addresses the critical issue of user-perceived quality of experience (QoE) in Internet video applications, which significantly impacts content provider revenues. Given the lack of support for optimizing QoE measures in the network, a robust bitrate adaptation algorithm in client-side players is essential to ensure a good user experience. The paper makes three key contributions: (1) developing a principled control-theoretic model to reason about a broad spectrum of strategies, (2) proposing a novel model predictive control (MPC) algorithm that optimally combines throughput and buffer occupancy information, and (3) presenting a practical implementation in a reference video player, dash.js, to validate the approach using realistic trace-driven emulations. The authors first formulate the video bitrate adaptation problem as a stochastic optimal control problem, defining key dynamic variables and an objective function. They identify the limitations of existing approaches that rely solely on rate- or buffer-based strategies and argue that combining both signals can lead to better performance. The MPC approach predicts the expected throughput for the next few chunks and uses this information to make optimal bitrate decisions for QoE maximization. To address the computational overhead and practical challenges, they develop a fast and low-overhead implementation, FastMPC, which uses a table enumeration approach to store optimal control decisions for future online use. The evaluation section compares the proposed approach against existing rate- and buffer-based algorithms using real and synthetic throughput variability traces. The results show that the proposed MPC approach consistently outperforms state-of-the-art adaptation algorithms by 15% in broadband datasets and 10% in cellular datasets in terms of median QoE. The FastMPC implementation adds negligible overhead to the baseline dash.js player, with only 60 kB extra memory usage and similar CPU usage.