News: are multi-head R/W HDDs finally on the horizon?

Hardware Hub Storage
1

While searching for news on the multi-actuartor HDSS, I’ve stumbled on this and it’s very fresh:

It seems that one of the reasons we haven’t had multi-head R/W so far was that existing second stage piezo-actuators weren’t precise enough to reliably track all heads at the same time while taking care of the invironment noise, vibrations etc.

SO now they seem to be about to introduce third-stage/micro-piezo actuators for ultrafine adjustments near the head itself.

if that brings per-head access, that’d be awesome. Especially of they implement full dual-stacks with two heads per surface.

Sure, power consumption would be somewhat higher, but transfer speeds would be just awesome.
In many scenarios where HDDs are just for cold storage or are bing cached by NVME, extra power wouldn’t be an issue.
And it would immenselly lessen many headaches, like RAID resilvering or initialization time, disk media surface test times and data backup/copying times etc .

Modern drives come close to 300MB/s per head. For a 11 platter drive with 22 heads that would be ~6.5GB/s per head stack or 13GB/s for dual-stack.

With 2:1 head redundancy redundancy per surface and and shorter average seek time.

Looks like next-gen HDDs wiill come with PCIe interface, just like existing NVME.
Which will FINALLY hopefully ground the prices and availability of PCIe switches. :roll_eyes:

EDIT: I wonder, with all that bandwidth and redundancy, if this means that new drives will also offer in-drive RAID-0/1/5/6/etc. Since all heads in a stack can access their sectors practically synchronously, this simplifies many things (buffering etc) immensely, so parity recalculation/data-integrity checks etc could be done in-drive. :roll_eyes:

2

We’ve had triple stage actuators on hdds for about half a decade so far, it’s an incremental improvement in tracking/density but I think TDMR is a much better improvement.

In order to implement multi actuator on a hdd, each head assembly would need it’s own voice coil since the micro piezo actuators don’t have enough range of motion to be able to access many tracks over to the inside/outside of where the VCA is nominally.

If I had to guess, I think only a small subset of HDDs made specifically for the hyper scalers will get the PCIe interface to help reduce their infrastructure variability.
The more normal segment of the server market will stick to SAS due to it’s higher performance/manageability/reliability.

3

I don’t think so. All tracks in a cylinder are already on a roughly the same position, sans mechanical tolerances. IMO piezo-actuators are exactly what is needed - just for microcorrections. Drive doesn’t really need the ability to move each head in a stack on independent track, it just needs to be able to keep all heads steadily over the same track X within the same cylinder so they can be engaged.

4

I see what you are saying.
I was thinking of implementing multi actuator in a way that would multiple the IOPS and throughput by however many actuators/platters the drive has, you’re talking about improving sequential throughput by just spanning the different different tracks within a cylinder and tying them all together as a kind of super-sector which is probably more plausible that what I’m thinking.

5

Yes, WIthout extra electronic, one could use all platter surfacess effectivelly as a mini-drives in a RAID-0 as a whole drive, with each head reading or writing its stripe.

With extra support it could be RAID-6 with spare surfaces for hot-spare or redundant redundancy (think of RAID-6+2) extra area for sector remapping or all of the above.