Next-generation metro access nodes with edge computing need to be redesigned to co-allocate advanced optical technologies, computing, and storage resources to support the upcoming multiple applications in 5G. In this paper, we present a novel metro access edge computing node based on a fast optical add-drop multiplexer with submicrosecond reconfiguration and control for low-latency operation. We investigate the network performance, the location of the edge computing nodes, and the computing resources dimensioning and utilization in order to fulfill the stringent latency requirement in 5G networks. Network function virtualization and network slicing have been considered in the model to emulate the realistic network operation. Optimization of the network, node location, and computing resources in terms of latency and packet loss ratio is numerically investigated via the OMNeT++ simulator under three different types of 5G applications (Massive Internet of Things, content delivery network, and loss-sensitive traffic). Considering a typical metro access network topology with 20 nodes covering a population of around 1 million, numerical results show that less than 200 μs latency is guaranteed for 5G network applications by deploying more than 6 edge computing nodes with 80 servers for each node.