It's Intolerably Slow (Sorry about the previous wording...)

1. Start a step-by-step decomposition of the performance numbers, by removing concurrency features and see how performance changes. Ideally after we have removed everything, i.e. CAS, mapping table, delta chain, SMO, it should be comparable to a single thread B+Tree using the same page size.

2. Using existing BwTree infrastructure to build a single thread B+Tree directly