Comment OpenIO, the French object-storage startup, is progressing its work with Kinetic drives.
The idea is that object storage stores data on server nodes, and those nodes could be individual Kinetic disk drives. Such drives have a micro-server, so to speak, running on them, providing direct Ethernet access and an object-style Get:Put interface. A slug of flash could be used for metadata.
Seagate's 4TB Kinetic drives used a Marvell Armada 370 SOC (System On Chip) controller. Marvell now has 64-bit ARM processor SOCs and OpenIO co-founder and CEO Laurent Denel sees an ecosystem developing, a set of small suppliers, including Chinese ones, that will take commodity disk drives and integrate a Marvell SOC to offer a Kinetic drive.
The Kinetic code is open source so it is now out in the wild, providing a generally available Ethernet bridge to SATA (or SAS) disk drives (or SSDs in the future?) across which object storage commands can pass.
This would mean that any small system integrator could produce Kinetic drives and not just Seagate.
So we might imagine 10TB drives plus Marvell Arm-based SOC running Linux and OpenIO software. And 12TB and larger capacity drives in the future. That's the implication of Denel's thinking.
OpenIO has joined the Kinetic Open Software Project. It can tier object data between SSD and disk drive and will add Kinetic drives as a third tier. The tiering data placement is driven by storage policies.
OpenIO can start an object storage setup with just three nodes, and scale from one object to billions as the system grows.
It seems to El Reg that having hundreds or thousands of Kinetic drive-based nodes is not going to be something we'll see overnight. Clearly scaling granularity will be much smaller if it's done at HDD-level instead of server (as in X86 server + disks/SSDs in a box) level. You could buy an empty JBOD enclosure and populate it at the rate you need as your object storage capacity responds to demand.
But, object storage is said to be good for large scale object numbers, billions and trillions of objects. If you do this with individual drives then you will need full-time disk-wranglers to load and plug in drives, and get the software working. Why not then get an object storage array and save all the repetitive work involved?
The mini eco-system Denel is seeing could have its own issues. For example, how will Kinetic system integrators be able to compete with Seagate's in-house kinetic drives on price? Seagate would be able to buy Marvell SOCs at far lower, volume-based, prices than any small Kinetic SI.
For us, Kinetic storage still has elements of clever engineering technology looking for an end-user problem to solve. Perhaps OpenIO has something hidden up its sleeve. ®