Re: Does AFA matter?
"There is a big difference in architecture and how systems handle flash when it's built from the ground up for that purpose."
Really? What part of it did they build from the ground up? Let's see:
1980s SCSI command set? Not until NVMe over fabrics is in the wild. Until then you're speaking SCSI.
Intel CPUs and DRAM? Nope. Same stuff in all the arrays except special ASICs from 3Par which are always late to the party... makes nice slides though.
Fibre Channel? Nope. Still works. Cisco and Brocade are pushing 32g FC for NVMe over Fabric.
Ethernet/FCoE/NFS/SMB? Nope. Not invented new for AFA.
Inline Compression/Dedupe? A lot of the AFA have those whether built from the ground up or not.
Scale out? Well that's not really an AFA feature and many of them fail that one.
Simple GUI or admin? Simplicity is nice until you need a feature. I agree some platforms are too complicated. It's a feature though not intrinsically tied to AFA.
SSDs or CFM? Most of them use SSDs and they should. My money's on SSDs. Going your own way on chips hasn't worked on for a long time. See Sun, DEC, SGI... Heck, the chipset landscape is cleared of competition. Where's Adaptec, AMI, LSI, National Semi...
Really it's just a storage array tuned for much faster backend access. It gives the manufacturers some new ability to tune their stacks instead of waiting for spinning rust to respond and lessen the need to buffer as much. There's great advances possible in AFAs but it's still just faster storage until we change the protocol. Even then, it's just faster. It's still retrieving the same old bits faster; not adding chocolate flavor to your bytes.
Does faster storage effect other things in the compute stack? Sure does. That still doesn't mean that it's fundamentally different from another array other than speed and with a neat bezel.