NVMe Polishes Its Specs, Brings Virtualization to Locally Attached SSDs (theregister.com) 22
The latest NVMe specifications add SSD-level virtualization that can simplify live VM migration by preserving storage identities across servers. The updates also introduce support for post-quantum cryptography, controller-based rate limiting, voltage monitoring, and factory-reset capabilities. The Register reports: Announced by the NVM Express consortium, all 11 of the suite of NVMe specs have been updated with new features and engineering change notices. These represent the next step in the evolution of the standard, it says, which was created as a protocol to support storage devices connected to a system's PCIe bus. Perhaps the most significant new capability is PCIe Exported NVM Subsystem Migration. This extends existing NVMe virtualization to locally-attached PCIe SSDs. It does this by abstracting the physical drives into host-defined virtualized NVM subsystems, to allow for virtual machine (VM) mobility without storage reconfiguration.
When a VM moves from one server to another, its storage also needs to move with it in a way that's non-disruptive to any applications running in that VM. "With NVM Subsystem Migration, NVMe SSDs can present exported NVM subsystems that hide the complexity of the underlying hardware," says Mike Allison, a senior director at SSD maker Samsung and NVM Express board member. "Instead of interacting with physical controllers and namespaces, the host only sees exported controllers and namespaces. This creates a clean separation between what the VM sees and what's happening under the hood," Allison explains on an NVM Express blog.
"One of the key innovations here is providing the host with control over exported identifiers. During migration, those identifiers can be carried over exactly from the source to the destination. That consistency is crucial: even if the underlying hardware uses different internal IDs, the VM sees no change and can pick up right where it left off with no storage reconfiguration required," he says. In effect, the NVMe layer now enables the virtual machine manager (VMM) to rely on virtualization built into the SSD. The flash drive itself exposes logical, virtualized storage constructs, offloading this complexity from the VMM. You can learn more about the updated NVMe specifications here.
When a VM moves from one server to another, its storage also needs to move with it in a way that's non-disruptive to any applications running in that VM. "With NVM Subsystem Migration, NVMe SSDs can present exported NVM subsystems that hide the complexity of the underlying hardware," says Mike Allison, a senior director at SSD maker Samsung and NVM Express board member. "Instead of interacting with physical controllers and namespaces, the host only sees exported controllers and namespaces. This creates a clean separation between what the VM sees and what's happening under the hood," Allison explains on an NVM Express blog.
"One of the key innovations here is providing the host with control over exported identifiers. During migration, those identifiers can be carried over exactly from the source to the destination. That consistency is crucial: even if the underlying hardware uses different internal IDs, the VM sees no change and can pick up right where it left off with no storage reconfiguration required," he says. In effect, the NVMe layer now enables the virtual machine manager (VMM) to rely on virtualization built into the SSD. The flash drive itself exposes logical, virtualized storage constructs, offloading this complexity from the VMM. You can learn more about the updated NVMe specifications here.
This is neat! (Score:3)
I want this feature like now. But to be really effective you need to be able to migrate between hosts as well. So the host needs to be able to set the device IDs, not the NVMe device.
I'm assuming that they've thought of that, though. Seems like they also introduced QoS. And what about using this for RAID? NVMe over TCP seems to be in there as well.
Re: (Score:2)
That is not a missing piece. That is pretty much the point of the exercise. The proposal lays it all out.
I managed Citrix and VMware pools backed by NetApp filers for decades. In that world, migration was comparatively straightforward because both hosts could already see the same backing storage. Direct-assigned local NVMe buy
Can't wait for SSD licensing (Score:3, Insightful)
SSD as a service! Can't fucking wait until I have to pay software licensing for my hardware.
Re: Can't wait for SSD licensing (Score:5, Funny)
Get off my lawn bro
Re: Can't wait for SSD licensing (Score:2, Funny)
Re: (Score:3)
They could have done this a year ago before the AI tax made CPU & storage spike from extra expensive to unaffordable.
Happy now?
Re: (Score:2)
This is Slashdot. Every story needs to have one AC yelling about how they are being repressed by ${story_topic_not_remotely_understood_by_poster}
Re: (Score:3)
but iSCSI, CEPH, ZFS, etc? (Score:2)
redundancy is big.
Re: (Score:3)
Most people would need that feature (Score:2)
Re: (Score:2)
If you run Windows 11, you run virtual machines. CredentialGuard is its own instance, and the Sandbox for web browsing is as well.
Overall, this isn't a bad thing, because more people should consider moving to virtualization anyway.
Re: (Score:2)
Complexity and failure modes :o (Score:2)
Features such as hardware-level virtualization, migration support, resource controls, and advanced management functions move more responsibility into SSD controllers and firmware, creating additional layers that must operate reliably.
Failures that were once limited to a single storage device could become more complex in shared environments, affecting multiple
Re: (Score:2)
Re: (Score:2)
A single server running in a rack has three failure modes. a) The hardware conks out b) the software borks and c) the power supply fails. Under the cloud, you have virtual machines running on virtual networks running on virtual switches. What could possibly go wrong
Re: (Score:2)
You provided three categories that are so broad that they never stop applying. That software emulates hardware changes exactly nothing under your definition. It's just more potentially failing software instead of potentially failing hardware. Which usually means "cheaper to fix".