The paper identifies the same trend we capitalized on in #LMDB: pure in-memory database designs are stupid. Workloads always grow to exceed the size of RAM.

This is especially true today as RAM prices continue to skyrocket. The smartest approach is a design that lends itself to being paged out to secondary storage.

And yes, the same problem we noted when designing LMDB: textbooks only show fixed-size records. Useless for real-world work.