There was a time when I knew a lot about the various storage technologies available for a database. It was important when designing a server around the various requirements for size and performance balanced against the limitations of hardware.
The rapid growth of solid state storage and the adoption of storage area networks have changed the game for many of us. We no longer care or think about storage. It's just a service that we consume in our databases, and while we might demand more IOPS capacity, we often don't worry too much about how that's provided to our systems.
I caught an article about IBM and a few enhancements to their offerings. They have a FlashSystem 5200, which is a way of taking their enterprise-class systems and packaging it in a way that organizations of all sizes can afford. In this case, they can get 1.7PB in a 1U package, which seems insane. I can remember working on an Enterprise 2U system that maxed out at 200GB worth of SCSI HDDs.
I wondered if this was a lot in today's world. A few friends work for Pure Storage, and I've met some customers who were very impressed with their arrays. When I looked at a FlashArray//X70 from them, they get 2.4PB in 3U, so the IBM offering is more dense. In terms of IOPS, I'm sure they are both very fast, likely fast enough for most places I've worked.
What's interesting is that I've found sysadmins way more concerned about specs and loving the Pure arrays more than the DBAs. DBAs (I think) are stuck with the hardware, and so they are always looking for what the bottlenecks are on a system and can I tune them down to a reasonable level. Or can I just get more storage or IOPs from whoever is in charge of storage?
The world of storage (and networking) has come so far in the last 30 years that the numbers today seem mind-bogglingly fast compared to even a decade ago. It makes me wonder why we have any performance problems at all. Unfortunately, all the hardware in the world can't compensate for keeping too much data in expensive storage and writing bad code. A little query expertise and archiving structures would likely make all that hardware look as amazing in the real world as it appears in spec sheets.
