For users of loosely-coupled distributed systems, questions about "performance" commonly involve interactive performance. The problem of measuring components of interactive response times is approached by representing the processing needed for each interaction as a graph called an interaction network. The critical path through an interaction network shows the sequence of steps that take the longest time to execute. The goal of improving performance is presented as requiring strategies to shorten the lengths of critical paths. To identify these strategies, we augment each interaction network with vertices that represent additional events of interest to a performance analyst. A particular aim has been a unified approach that treats all types of system resource (hardware, software, or data) in a consistent fashion. An example is given to show that a set of event recorders can be designed to provide measurements for an exhaustive decomposition of response times for selected sets of interactions. Finally, details are given of experimental work done to record and analyse interaction networks.