back to article NVMe fabrics could shuffle traditional arrays off to the graveyard

NVMe-over-Fabrics (NVMeF) shared storage access could kill the legacy storage array business – unless vendors get inventive and somehow continue to supply charged-for data management services alongside NVMeF data access. How do we work this out? An NVMeF setup works with applications in servers requesting a storage IO, and …

  1. Duncan Macdonald

    Raid inside SSDs ?

    As there has to be a controller inside each SSD to handle the mechanics of block allocation, wear leveling etc, it would be feasible to have RAID 1 logic inside this controller with 2 flash arrays on the SSD. (Higher RAID levels would probably not be feasible due to the increased complexity and overheads.)

    1. The Mole

      Re: Raid inside SSDs ?

      You could have raid built into the SSD, but how useful that is depends on what you are trying to protect again and what failures you are expecting. With SSDs I wouldn't expect that the failure is on the individual NVM chips but more likely to be the auxiliary components on the device - e.g. the external interface, diodes and resistors. Failure of these is likely to take out the entire device making useless the redundantly stored copies. Even with a good degree of separation most failures are caused by something, such as localized heating or power surges, within the same device there is a real chance the cause will damage both circuits at the same time.

      Independent units are far more likely not to fail synchronously, and given that RAID1 halves your storage, you are going to need twice the number of devices anyway so you may as well keep the RAID between the disks instead.

  2. Field Commander A9

    So...Where does this leave HCI?

  3. Pascal Monett Silver badge

    Is this a case of having your cake and eating it too ?

    From what I gather, there is a move to use hardware to make data transfer requests quicker. I get that, but if I'm not mistaken there is a global move to have network supervising and management via software - and that includes actually defining the network via software.

    Seems to me that these two movements are in opposition with each other. You can't have a software-defined network that relies on hardware-coded optimizations.

    Or can you ?

  4. ntevanza


    Cars are a lot faster than they used to be, and there are more of them about. Here is a list of 'data management' feature your car probably has. You've probably used some of them without even realizing it:

    Brake servo

    Adaptive power steering

    Adaptive cruise control, adjustable, with various warnings

    Electronic throttle (or more likely, fuel injection programming), for the

    Stability control

    Traction control


    Automatic braking after a collision to prevent run-on collisions

    Tyre pressure sensors


    Seatbelt pretensioners

    Lane detection with or without steering input

    Blindspot object detection

    Parking sensors

    Ice warnings

    Driver drowsiness detection

    Intrusion detection & immobilization

    Automatic hazard lights and emergency calls after collisions

    If you do get a fast car without all these features, or you want to turn them off, then drive on the track, aka test/dev, where you can't hit anything. Track driving is a legitimate use case. It is also specialized and unusual.

    Your OS has maybe two of these features. Your apps are drunk, libidinous teenagers who want to borrow the car. You figure it out.

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Other stories you might like