Very much not an expert, but I think this is talking about a whole different technology. This is holographic storage within a 3-D medium.
The article explains that the technology is 20+ years old, but didn't seem to occupy an optimal point on the frontier of throughput and cost, which is the major tradeoff for long-term storage devices. But it says with the physical limits of spinning magnetic disks on the horizon, it's worth looking at holographic again.
It further explains that to characterize the cost, you have to consider that the data has to be refreshed periodically, because reading degrades it, so that must be considered in the cost. And then they describe a protocol for doing the reads and refreshes that has good results.
It's not at all the same thing, but I always wished Project Silica[1] would take off. I want durable, scalable write-once storage that I can put in a file cabinet for decades and then reliably read.
The end of this article they claim their tech can do 4 IOPS/watt. Thats the theoretical power consumption, not including any computation to decode the data etc.
I assume that one 'IO' means reading a 47 kilobyte page. That means reading at say 200 Megabytes per second (like a hard drive) would take 1 Kilowatt!
They also say a regular HDD has 20 IOPS/Watt. Consider the WD HC570[1] which can do around 220 IOPS in mixed 4KB workloads, and consumes around 9W doing so.
So this seems what they mean, and thus they're "only" a factor of five or six away from the efficiency of a modern HDD.
I think the difference is that a hard drive can read sequential pages far quicker, and at far less energy per byte, whereas as far as I can see this tech cannot. Random and sequential reads would work out the same energy per byte.
Right, to put some numbers to that the same HC570 can do peak 291MB/s which divided by 4KB pages means 74496 sequential IOPS. Hence why we love SSDs.
However to their defense, they seem to be quite focused on this being for cloud vendors. And, as has been demonstrated[1], a sufficiently large number of sequential IOs effectively looks like random IOs to the the server.
That being said, surely this is more for cold data, so I'd imagine the cloud vendors can optimize access to benefit from the high sequential IO bandwidth of HDDs.
I doubt it. There were a number of attempts to create commercial holographic storage formats (primarily optical discs) in the early-mid 2000s, but none of them panned out. At this point, any new technology would need to show substantial advantages over flash memory and/or magnetic hard disks to be commercially useful; that's a very high bar to clear.
I remember hearing about it at the end of 90s in a paper computer magazine. What's different this time?