This thesis describes a study of load balancing and topology discovery in the Internet. Topology data was collected using variants of Traceroute including Traceroute MDA which provides topology information about load balancers. Optimised Traceroute MDA was applied to investigating or updating basic statistics describing the behaviour and performance of load balancers and their populations in the Internet. Simulation studies were then performed to optimise the efficiency of distributed topology discovery algorithms including an algorithm for discovering load balancers, as well as one based on classic Traceroute. Initial optimisation of the load balancer discovery tool was based on refining the parameters used to decide when probing should stop at each hop along the Internet path under analysis (stopping values). Internet data was then collected using the tool to attempt to confirm the refined parameters in practice. Large repeated analyses were performed to quantify load balancer population parameters and change over time. These results included changing prevalence, turnover, load balancer diamond structure including a new statistic called literal width, fields outside the classic five-tuple that control load balancing and flow ID selection factors that affect the efficiency of discovering load balancers. Data was also collected on the occurrence of black holes in load balancers. The classic Traceroute based tool Doubletree was simulated to test and improve efficiency, as was our new load balancer topology discovery tool Megatree. Doubletree was simulated with both a simple and advanced simulator and a simple simulator was used for Megatree. These topology discovery tools store information within vantage points and communicate between vantage points to reduce traffic required for topology discovery. New approaches to using these tools helped improve efficiency when large numbers of vantage points were used.