Better late than never
Although SCM on the array is literally "lipstick on a pig". The real value of SCM is on the server, not the array when it comes to high performance and lower latency.
Dell EMC has injected FC-NVMe and Optane storage-class memory (SCM) into its high-end PowerMax arrays with the aim of ramping performance. The updated PowerMax specs are up to 7.5m IOPS, sub-100μs read latency and 1PB effective capacity for the 2000, and up to 15m IOPS, sub-100μs read latency, 350GB/sec and 4PB effective …
If I’m paying a premium for SCM, it would make sense that I would want all my hosts to be able to take advantage of it as well as protect my data with data services like snaps, replication, and encryption. Plus if I lose the host with SCM in it, I might lose the data and no one else can access that SCM. Sticking SCM directly into servers is so FusionIO-ish. That didn’t work out too well for them.
Agree on a single server without any other services. But consider the use-case for shared database cluster disks. Shared hypervisor storage, Vmware’s vmotion. PowerMax’s Storage Direct Backup capability, array based DR Replication in SRDF; DR without stealing host CPU or Host Bus Adapter cycles. Active/Active support with SRDF/Metro, etc.
Yes I am a 25 year unabashed Symmtrix/DMX/VMAX/PowerMax fanboy.
Any impartial infrastructure nerd will tell you that SCM needs to be as close as possible to the server CPU to have the most value, where latency is measured in nanoseconds, no microseconds.
Adding into a SAN doesn't really accomplish much for an application running in the server - latency will still exist between SAN and server, no matter how fast the SAN can respond to data block request.
It may have some limited use in some of the SAN data services - mirroring, replication, some of the inline data activities (deduplication, data integrity checking, etc) but with remote replication, link aggregate bandwidth will be the limiting factor here.