Throughput-optimal network controllers employ dynamic methods to fully utilize the network resources in terms of achieving optimal long-term throughput performance. However, in the process of dynamically searching for the optimal paths and rates, these controllers typically result in many redundant transmissions, loops, and unfair distribution of end-to-end delay amongst packets that belong to the same session. In this work, we explore several novel and promising techniques that aim to prevent these vital deficiencies without sacrificing significantly from long-term performance guarantees. Further, our approach extends the current state-of-art by developing a new architecture that utilizes a virtual layer algorithm in the real-world implementation to achieve both throughput-optimality and significant delay performance improvement over existing strategies.