Linux 6.13 Rolling Out NVMe 2.1 Support & NVMe Rotational Media

[billyswong]

i know about U.2 but thats a 2,5" form factor, and while such HDDs exist, they dont in capacities where its capacities where SSDs arent just better in every aspect

I think there isn't any restriction in U.2 drive form factor? HDD vendors definitely can manufacture a 3.5" drive with U.2. Or whatever any other sizes. The device plug specification only imposes a restriction in minimum physical size, not a maximum.

[elvis]

as for the 2nd, introducing a compatibility breaking change to a dying technology isnt exactly the best idea

Tiered and hierarchical storage are still MASSIVE markets. I'm working for multiple customers who have single-digit-petabytes flash, double-digits-petabytes spindle, and who-knows-how-much tape, and very large management tools that shuffle that data back and forth for end users.

In the same way that tape has been "dying" for decades, spindle continues to offer a great cost-point for middle tier storage at huge volumes. And again, when you start talking about very large JBOD storage systems (including "zero watt" systems for object storage where individual disks can spin up and down on demand).

And in terms of commercial volume - these are the biggest hyperscalers. AWS, Azure and GCP are all still buying spindle at enormous scale (and tape too!).

Looking at consumer devices as a metric for what technology is "dying" is the wrong way to go about things in late 2024. Commercial viability is now very much in the hands of the huge players.

[davidbepo]

Tiered and hierarchical storage are still MASSIVE markets. I'm working for multiple customers who have single-digit-petabytes flash, double-digits-petabytes spindle, and who-knows-how-much tape, and very large management tools that shuffle that data back and forth for end users.

In the same way that tape has been "dying" for decades, spindle continues to offer a great cost-point for middle tier storage at huge volumes. And again, when you start talking about very large JBOD storage systems (including "zero watt" systems for object storage where individual disks can spin up and down on demand).

And in terms of commercial volume - these are the biggest hyperscalers. AWS, Azure and GCP are all still buying spindle at enormous scale (and tape too!).

Looking at consumer devices as a metric for what technology is "dying" is the wrong way to go about things in late 2024. Commercial viability is now very much in the hands of the huge players.

interesting, now do notice i said dying, not dead, i know about tape still existing on the ultra high capacity level, but HDDs will eventually get squeezed from the bottom by SSDs and from the top by tape, since both are inherently simpler and cheaper techs
i dont know what the HDD/tape crossover is, but SSD/HDD one is 1TB and its only gonna get higher as flash tech gets cheaper
also important to note on HDDs dying, while way more expensive, biggest SSDs already have more capacity than biggest HDDs, so it is PURELY a price thing, as SSDs will give higher capacity and WAY higher capacity density(2,5" vs 3,5")

further info on this, biggest HDD i could find is 24TB, 3,5", biggest SSD is 61TB 2,5" U.2, so difference is actually bigger than i anticipated

anyways i reaffirm HDD is a dying tech, but i guess it has long enough where this may be worth it, if only for hyperscalers

[elvis]

further info on this, biggest HDD i could find is 24TB, 3,5", biggest SSD is 61TB 2,5" U.2, so difference is actually bigger than i anticipated

Seagate has a 32TB available now:

https://www.seagate.com/au/en/products/enterprise-drives/exos-x/x-mozaic/

WD announced 40TB soon:

https://www.marketbeat.com/earnings/transcripts/98427/

These are SMR, but again the application here is mostly for "nearline storage" - i.e.: long term store files, S3 objects, backup data, etc, but with faster access / lower recall latencies than tape. SMR, like spindle, is constantly called out as invalid technology, but the reality is that there's massive commercial demand for this because of its price-to-size ratios.

Yes, flash media is beginning to exceed spindle in terms of density and maximum size per drive, but the price of those huge flash devices is often an order or two in magnitude larger. Hyperscaler problems are multi-dimensional - how many petabytes can you squeeze into a given space for a given dollar cost and a given wattage draw?

Flash is absolutely gaining ground here, but spindle still exists precisely because it's cheaper with respect to these specific problems. When those ratios flip, I have no idea. Every time someone announces the upper bounds of what spindle can achieve, some research group somewhere proves them wrong with another bump in density. Who knows where that will end.

anyways i reaffirm HDD is a dying tech, but i guess it has long enough where this may be worth it, if only for hyperscalers

Everything is "dying tech" in the big picture. Spindles will endure as long as they meet all of the price/size/performance/power/density/cost ratios required by their customers. And "customers" look a lot different today, with the explosion of stuff in data centers. "Dying" is always relative, and often non-linear (vinyl records are seeing a resurgence - sometimes old things come back for strange reasons).

I've spent close to 30 years in this career, and I've learned several times over never to count a particular technology out. Even when you think you've seen the last of it, someone somewhere will come out of the woodwork with enough commercial demand to see things revived back into production life. See languages like COBOL and FORTRAN, architecture like mainframes and IBM POWER, or technology like Infiniband and RDMA. Every time someone announces the death of these, I get another contract keeping these things alive another decade for someone, somewhere, often with lots of expensive commercial support from modern vendors.