NVMe storage¶

General concept¶

NVMe based storage is the fastest available persistent storage in the platform. This is due to:

• NVMe uses flash (SSD) cells connected close to the CPU of the computer (on the PCI Express bus directly - similar, but not identical, to RAM).

• NVMe storage in Binero cloud is local to the hypervisor, meaning networking is not used to reach the storage.

• NVMe storage is NOT replicated, meaning there is no added latency when writing data.

These attributes combine to dramatically lower the latency from request to response (for both reads and writes) to the storage layer.

NVMe storage measures in microseconds as opposed to other storage that is generally measured in milliseconds. As latency is arguably the most important factor in storage performance, reducing it will certainly have an impact on storage intensive applications where CPU will not have to idle while waiting for IO as much.

How to use¶

When provisioning an instance that is including NVMe in its name (for more information, see our list of flavors), the platform will schedule the instance onto hypervisors with local NVMe disks attached.

With NVMe, you are only able to select between 50 GB or 250 GB disk sizes, the reason for this is because NVMe disks by design are local and does not provide the same flexibility as our block storage service.

You are not able to grow a disk or retype it either for the same reason. Because of this, we recommend choosing an NVMe disk that fits your need.

Differences to standard storage¶

Since NVMe-based storage is setup differently than our standard SSD and HDD storage types, there are some caveats that a consumer of NVMe storage should be aware of:

• NVMe storage is NOT redundantly setup. We deliver NVMe storage from the same hypervisor that runs the instance to provide the performance that NVMe disks does, they are not using RAID nor are they replicated. That said, they are high enterprise grade disks.

• NVMe storage cannot be live migrated. Live migration is a technique that our administrators use when doing system maintenance to evacuate parts of the platform to allow for maintenance with zero downtime. Since NVMe uses a local resource, its not possible to live migrate. Because of this, when doing maintenance, instances based on NVMe is powered off for the duration of the maintenance. Please see our Outages article that details how to get noticed in case of planned maintenance.

• NVMe-based storage cannot be extended through the platform, if you think you will ever need more than 50 GB we recommend going with 250 GB. If you need more than 250 GB, contact us to discuss a solution.

• NVMe-based storage cannot be retyped. You can add one or more extra HDD or SSD volumes to the same instance and copy the data. Since the NVMe volume is the boot disk, we recommend doing a migration to another instance if you need to retype.

• Since we need to write images to the local NVMe disk on the hypervisor, it takes longer to provision an NVMe based instance.

• NVMe based storage is currently only available in the europe-se-1a availability zone.

• Because NVMe is generally used only when performance is critical to the application, we only have our High Performance Flavors available with NVMe.