What should the solutions architect recommend to solve these issues?

Explanations:

While Amazon Aurora with Aurora Replicas can improve database performance through read scaling and high availability, it may not specifically address the requirement for sub-millisecond response times and snapshot replication in a manner that matches the needs of a high-demand mobile game. The latency improvements would still depend on the underlying RDS performance, which is currently identified as a bottleneck.

Amazon DynamoDB with global tables offers high availability and low latency, but it is a NoSQL database that might not support all use cases of the existing RDS database, especially if complex queries or transactions are necessary. Additionally, while it provides replication, the response times may vary based on the workload and table design, and it does not inherently guarantee sub-millisecond response times.

Adding Amazon ElastiCache for Redis would provide an in-memory caching layer that can significantly improve read performance and reduce load times for frequently accessed metadata. Redis is designed for low-latency, high-throughput operations, allowing for sub-millisecond response times. It also supports data persistence, which can help with snapshot capabilities, and can be set up to replicate data across multiple nodes for fault tolerance.

While Amazon ElastiCache for Memcached can also provide an in-memory caching solution, it lacks some advanced features like data persistence and replication that are available in Redis. Memcached is optimized for speed and simplicity but may not meet all the requirements for snapshot replication or maintaining state across multiple instances as effectively as Redis can.