Uber has recently faced significant performance issues due to increased traffic, revealing memory-related bottlenecks like long GC pauses and memory leaks. Their engineering team documented their optimization process, including identifying a misconfigured thread causing excessive object creation. They utilized tools like GCeasy and fastThread for effective analysis and resolution of these challenges.
The decision to use few large memory instances versus many small memory instances lacks a definitive answer, as evidenced by two successful technology companies adopting opposite strategies. Both approaches can work if backed by a competent team. However, large memory configurations are typically more expensive and can complicate performance troubleshooting.