"And what about cache coherency?"
The simplest example is a DBMS - where you currently have a large database that's much bigger than available RAM and you rely on SSDs for performance, you would now move to holding the entire DB in PMEM (assuming cost is similar to SSD but still much less than RAM) and either reducing RAM or gaining additional performance while using cheaper storage for your long term "permanent" storage needs (i.e. spinning rust or a disk/tape backup solution). All of your cache coherency issues are already managed by the DBMS so there should be minimal changes required with the exception of optimisations.
For other applications, the degree of difficulty in implementing PMEM and the subsequent benefits are left to the individual to judge as not everyone needs the additional performance.
Biting the hand that feeds IT
