Let's wait and see what it costs when it comes to market and when.  It looks like great persistent storage, but fuel cells have been just around the corner for almosy 50 years now.

On Thursday, August 6, 2015, 'Leyne, Sean' Sean@... [Firebird-Architect] <Firebird-Architect@yahoogroups.com> wrote:

 

Jim,

> One, of course, is persistence.

XPoint memory claims the same persistence characteristics of SSD/NAND Flash, but with a much higher level of endurance.


> The third is whether reliance on RAM forces a very expensive scale up growth path.

That could be a concern.

If we look at current DDR4 16GB ECC DIMM prices ($140 USD ea), given the 10x increase in density. It could be reasonable for 160GB of XPoint memory to be available for $500.

So, spending $3,000 USD for 1TB of XPoint memory would hardly be considered a major impediment for scale-up.

We'll see, maybe in our lifetime.  Keep in mind that is probably isn't drop in replacement for RAM.

> This is where the distrubuted "atom" architecture of NuoDB and Amorphous came from.

Distributed has its uses, but a single system could be much easier to deploy and manage. What is the ‘cost’ of the distributed coordination? How much time are CPUs actually waiting for data from disk?

Redundancy, fault tolerance, and scalability all argue for distrubuted solutions.

Unless no node has the atom it memory, a request never waits for a disk read. 

 

 

--
Jim Starkey