Scaling a system to handle 100,000 TPS (Transactions Per Second) is rarely solved by simply adding more database replicas. The bottleneck almost always shifts to disk I/O latency. Implementing a dis…