I ran some tests today. Without caching, I got 5.36 transactions per second (tps). If commit history is cached, I get 7.25 tps. What actually happens in a transaction isn’t relevent. What is important is that by caching I get 35% more transaction per second. That is, the system performance increase by 35%. That’s pretty significiant.
Next step: cache the commit details and see how much faster that makes the system.
