After their recently launched technology, WD claims it will be able to bring a 40TB drive to the market by 2025. I went through their announcement and a few articles pertaining to it, but it’s still unclear to me if these drives will be able to handle a higher number of IOPS or throughput compared to today’s standards (2025 is indeed a long time to wait). Anyways, such a drive would become a massive failure domain, with recovery times that could become unsustainable. Even if it’s somewhat faster than tape, the IOPS/GB is not impressive.
Tapes on the rise (once again)
At the same time IBM launched a 30TB tape drive. And if the (very small now) tape industry manages to maintain its promises, we’ll be seeing huge tapes in the years to come. These tapes are the slowest media, but are also the cheapest and easier to store for decades-long cold storage.
Flash is unstoppable
The same goes for Flash. 3D NAND, TLC and QLC… all technologies that are helping to squeeze more data per chip and there is still a lot of room for improvement. With vendors already talking about 30/60/128TB drives and 1PB/1RU, depending on the form factor, the number of use cases for Hard Disks will be narrowed down pretty quickly. Yes, we can still argue about $/GB but it’s also true that a good part of this issue will fade away with the investments that flash vendors are making on new fabs and production processes.
Closing the circle
If vendors don’t find a way to significantly improve performance of next generation hard drives, I can’t see how they’ll be able to keep up with tapes for cold storage or flash in all other use cases. And I’m not just talking about the ability to access data quickly, (maybe through a large flash-based cache for random IO) but throughput should improve too. They might be able to find a small niche for active archives or cold-but-no-so-cold storage? I don’t know, but it will become increasingly harder to justify the existence of hard drives in the datacenter if someone doesn’t come up with an idea to improve them. And again, the failure domain will be outrageously huge if these disks are installed in large x86 servers with 60, or more, slots and resiliency is again in favour of flash and tape