Network infrastructure is composed of various devices located either in the core or at the edges of a wide-area network. These devices are required to deliver high transaction throughput where a transaction may involve processing one protocol data unit (PDU). Throughput of network infrastructure applications running on generalpurpose architecture based servers is constrained due to excessive memory access latencies and limited memory transfer bandwidth. In this paper, we analyze the memory access characteristics of three network infrastructure applications: IP forwarding, HTTP proxying, and RTP streaming. In addition, we analyze the latencies of these network applications with respect to three types of data transfers: memory-to-CPU, memory-to-memory, and memory-to-network. We calculate the optimistic upperbounds on throughput for these applications on generalpurpose computing platforms.