The number of possible aggregations and data mining approaches is probably only limited by engineers' ingenuity. One very common and relatively easy-to-implement approach is "feature extraction." It refers to a process that takes a full trace and calculates one or more values, called features, that are otherwise not possible to compute from a single span. Feature extraction represents a significant reduction in the complexity of the data because instead of dealing with a large directed acyclic graph (DAG) of spans, we reduce it to a single sparse record per trace, with columns representing different features. Here are some examples of the trace features:
Total latency of the trace
Trace start time
Number of spans
Number of network calls
Root service (entry point) and its endpoint name
Type of client (Android app or iOS app)
Breakdown of latency: CDN, backend, network, storage, and so on
Various metadata: Country of origin of the client call, data center handling the request, and...