The market is saturated with solutions for middle-boxes that make hard drives talk to networks, but nobody seems to be directly addressing the problem of we just want storage network accessible.
It's my experience that these boxes try to do a hell of a lot more than just putting drives-on-network and that is why they all suck and are expensive.
The NVMe-oF fabric devices out there all seem to command a ridiculous premium when the reality is they ought to be very simple and easily cost-optimized.
ATAoE ("aoe" in the Linux kernel) is nice because it is very lightweight in both terms of code to implement it (~2-3kloc, basically just stuff an ATA packet in an ethernet frame), low network overhead, and ease to setup (no IP addresses).
Nothing, because it makes each disk quite costly and by the 2014 nobody wanted costly and slow HDDs.
Check the Seagate offering up there it has 1Gbit interface. You can't even run the drive at full sequential read/write speed over it. And having a two 10Gbit ports on each drive would require having two 10Gbit switches, which by 2014 were still quite costly.
EM6 solution[0,1] is neat but at least it both the quite packed both by the spec and the price but delivers a lot of IOPS and throughput.
[0] https://www.ingrasys.com/assets/files/Datasheet_ES2000_20211... [1] https://www.servethehome.com/ethernet-ssds-hands-on-with-the...
Most hard drives run a serial console on some of the jumpers. You can easily run PPP or SLIP over that. QED :P
BTW if you already explored that, would you know how to alter the SLC/MLC ratio by any chance?
Modern QLC drive often have a SLC area for buffering. With the right firmware tools, it should be possible to take a 4 TB QLC drive and convert it to a 1 SLC drive to get more performance.
There were some firmware bugs on Seagate Barracuda SATA drives that could be worked-around w/ serial console. I don't remember the specifics though.
If you some search-engining on hard drive manufacturers and "serial console" you'll find indexed pages. (Presumably the really interesting stuff is buried deep down in forum posts, etc.) Just doing a couple quick searches got me some stuff.
Well, iSCSI has been around for ages, and because people got fed up with Fibre Channel requiring dedicated switches and transceiver components, first came FCoE that allowed using regular network transceivers and switches and then FCIP/iFCP that added regular IP routing to the mix but never saw much uptake.
From rough memory, didn't FCoE start out that way only in theory?
With the reality that people needed to buy FCoE rated equipment, which was priced at "enterprise pricing" levels.
Pretty sure that was FCoE, and I'm not mis-remembering that from something else... :)