Memory exhaustion during large-scale graph instantiation in NetworkX

NetworkX utilizes a dictionary-of-dictionaries data structure to manage nodes and edges, allowing for flexible attribute storage. While this provides versatility, the memory overhead per element is significantly higher than in contiguous array-based structures.

When scaling to extremely dense graphs or datasets with millions of edges, the Python process may encounter memory exhaustion before the graph is fully instantiated. This is particularly critical when using MultiGraph or MultiDiGraph, where parallel edges further increase the dictionary depth and memory footprint.

Given the current implementation of graph storage in NetworkX, what are the recommended strategies for monitoring memory growth during the construction of dense graphs? Are there documented configuration adjustments to reduce the per-node memory overhead without migrating to a different library?