Design a high-level architecture for a globally distributed social media backend that must handle 100 million concurrent users with sub-100ms latency for post retrieval. Explicitly address how you will handle the following: Data sharding strategies, consistency vs. availability trade-offs (CAP theorem), handling network partitions, edge caching strategies, and database selection (SQL vs. NoSQL). Provide a diagram description of the components and explain the data flow for a 'Write' operation and a 'Read' operation in detail.