Seagate has been pursuing dual-actuator drives because density increases have been causing a decline in performance per TB, making the performance characteristics of big hard drives relatively more tape-like. It's a first small step in the direction of your proposals, but it doesn't seem to be a big hit so far. That indicates taking those ideas to their logical extremes is probably very uneconomical.
Speaking of tape like behavior, it would seem that SMR would benefit a whole bunch from multiple actuators. It seems like rather than having dual actuators, as that doubles the number of simultaneous streams, it would help with lots of shared workload, but really that problem should be solved somewhere else -OR- widening the read head by some multiple. Instead of using a record player needle, use a rake. Interleave the tongues so there are like k tracks between splines. Also reduces the actuator travel distance.
We should think of individual drives not as DIMMs but the individual chips themselves. They can be remixed into what ever your workload demands.
Consumers will be super excited that their desktop computers now take twice as much space.
Why build bigger drives if you can do the same with the existing drives, RAID and SSD caching?
https://diskprices.com/?locale=us&condition=new,used&units=g...
You are all right and much smarter than me. I kinda outlined that lots of aggregated 3.5" (9cm) drives are better than some sorta boomer 30cm platters. I'll admit it was kinda for aesthetics.
1. increase their device bandwidth SAS only does what, 1200MB/s ? Networks are at 200-400G right now. And an individual SAS device can only do 12Gbits.
2. take predicates, even if the interface can handle 4-16GB/s your internal bandwidth and the ability to take a predicate means the drive can start understanding the data it stores
3. HDD physical media bandwidth needs to increase. Hard drives should move to 1-4U sizes, spin slower, have massively larger platters and multiple read heads.
If we assume 200GB/cm^2, we come out to ballparkish numbers for modern drives if we assume 5 platters. Lets just say that is true.
A single side of a 30cm platter is equivalent to 5 platters, two sides so 10 surfaces of a 9cm platter. Lets ignore the space taken up by the hub.
So a 19inch rack sized storage device is going to have 10x the storage capacity. It might have 10x the HD heads? Is more a chicken all in a single egg and ... I think would have to do something more impressive to get economies of scale.
The current 3.5" drive is going to pack a lot more than 10x to the 1u. So I think the current target I'd like to see drives go is a higher bandwidth access to the device, larger buffers, ability to send search predicates and a higher level protocol. We need HDD with NVMe. One should be able to do 4GB/s to an HDD for at least 10 seconds. All HDDs should be hybrid except for DVR drives as they only see there sustained write rate.