The present Internet is used to support a diversity of application types, each with different performance expectations. Latency is a fundamental component that affects end-user's Internet experience. Many applications can benefit from reduced latency. The thesis examines the startup latency of the Standard Transmission Control Protocol (TCP Reno) congestion control algorithm when used for short-lived interactive applications. TCP plays a key role in preventing Internet failure by decreasing network utilization when it detects packet drops or receives explicit congestion signals. TCP Reno senders react to congestion by reducing their rate by approximately half. When a flow has recently started, TCP Reno has no way to determine the available capacity, and uses a conservative startup method that can result in high latency for short-lived interactive applications with only a small amount of data to send. This is especially so when there is loss of the connection setup packet before a flow starts. This thesis therefore proposes a new algorithm called TCP SYN Loss (TSL) Startup Algorithm that uses a halving congestion response function during startup, which is less conservative than the current method. After connection setup is completed, standard TCP congestion control is applied for the data transfer phase. The new method is shown to significantly reduce web response time by increasing TCP Reno robustness against startup loss. Startup latency also affects users who rely on broadband satellite for Internet access. Application performance is impacted by the delay of bandwidth-on-demand (BoD) multiple access control (MAC) protocol at the start of a TCP flow when there are no prior capacity assignments. This thesis proposes a new method that combines Random Access (RA) and BoD to reduce web response time and startup delay of media streaming over satellite. Furthermore, a modified algorithm called the Split RADAMA protocol was designed to avoid the negative transport effects when switching between RA and BoD transmissions e.g. packet reordering and spurious retransmission. Together the proposed methods can significantly improve web experience for a wide range of end-users by reducing the response time for short-lived interactive applications.