Bandwidth management and flow control for Internet backbones

With the wide deployment of ATM, we may observe the flourishing of many virtual private and public Internet backbones configured over the same pool of ATM resources. Currently, these backbones are likely to use the CBR service, with resource commitment to each customer and perfect traffic isolation across different backbones. However, the CBR service is expensive and effective resource allocation and utilization becomes of paramount importance.; In this dissertation I study the alternative use of ABR on Internet backbones. Some resource commitment is provided by the MCR negotiated for each ABR connection, but the Internet traffic can use added resources because the ABR flow control recovers the unused bandwidth available along the paths of ABR connections. In addition, I combine the ABR backbone features with CBQ scheduling to implement dynamic MCR re-negotiation as a means to tune the allocated resources to the offered backbone traffic.; While the ABR flow control improves ATM resource utilization, the bandwidth leftover by very bursty background VBR traffic is not entirely recovered. This is due to the inability of the ABR flow control to track the variations of the aggregate VBR traffic. In this dissertation I study the limitations of the current ABR flow control standard to realize perfect recovery of available bandwidth and I propose a bandwidth over-booking scheme to address this issue.; Finally, the bulk of the Internet Best-Effort traffic uses TCP, whose window flow control is known not to perform well if a very large number of flows are simultaneously active. I present an analysis to study the TCP dynamics in this scenario to highlight its aggressive behavior under many flows. Finally, I propose a local max-min fair back-pressure flow control mechanism for ABR backbones that improves the TCP performance.