The Internet today has emerged as a ubiquitous network consisting of globally-shared resources. Optimal sharing of these resources raises the issue of resource and cost allocation which, in turn, leads to network performance modelling. In our work we emphasize the importance of performance evaluation and monitoring of network
resources to achieve their optimal utilization. We analyze the network as a decoupled system consisting of end nodes and routers. We analyze the algorithms running on each component and propose modifications. For the router we study the existing and widely-deployed active queue management scheme, Random Early Detection (RED) and its predecessor, TailDrop scheme. Emphasizing
robustness and end-to-end delay, we propose our modification to RED and show that it achieves better results compared to RED. Similarly, for the end nodes, we explore the window adaptation
scheme of the widely-deployed cooperative transport protocol, TCP Reno. Stressing packet re-sent ratio and power, we show that our
proposed modification of window adaptation schemes achieves better performance than TCP Reno. For a more balanced research we compute
the algorithmic complexity of each algorithm to show that better results can be achieved at the expense of increased algorithmic complexity.
