
The point of all flash arrays was not so much speed but guaranteed consistent low latency. Even hybrid arrays were able to occasionally deliver 1-2ms latency but would generally be 3-4ms and (very often) go up to 10 or more. Spikes of 20+ ms are common and source of many worries.
AFA basically guaranteed (sub)ms latency consistently.
While this is a nice lab test (although I loled at iometer), it gives barely more than bragging rights until someone else tests with NVMe and Optane. Doubt that Scale's IO subsystem is better than Datrium.
The name of the game is performance consistency. I want my Oracle and MSSQL system to scream at 10microseconds ALL THE TIME, not for some arbitrary queries and be slow for others.