LINSTOR

Persistent Container Storage

With the LINSTOR-CSI driver, the storage volumes provided and managed by LINSTOR integrate seamlessly with Kubernetes and the workloads running there. In order to meet the persistence guarantees, it allows the user to select the replication count per Storage Class.

Block Storage Replication

Data replication is necessary to meet persistence guarantees in a world where hardware devices might fail. The highest performance is achievable when this replication is done at a very low level, at the block level below the file system.

Kubernetes Integration

For Integration with Kubernetes, all components are packages as docker containers; DRBD-kernel-driver module loader, LINSTOR controller, LINSTOR satellite, the LINSTOR CSI driver and the LINSTOR Operator.

Separation of Data and Control Plane

LINSTOR is only necessary for control operations like; Creating volumes, taking snapshots, resizing volumes, or making them available on other nodes. After a volume is set up and handed over to the workload the IO operations on it are independent of the LINSTOR controller and LINSTOR satellite components. Those could be restarted/upgraded without affecting IO on existing volumes.

RDMA

As an alternative to TCP/IP over Ethernet DRBD can use RDMA networks natively. They are RoCE and Infiniband fabrics. That frees CPU for the workload itself, enables bleeding edge network speeds and lowest storage access latencies.

Management of PMEM

PMEM is a new and very fast storage class. When it is available in addition to SSDs LINSTOR can use it to place DRBD’s meta-data and add dm-cachen & dm-writecache to accelerate IO throughput and access times.

Snapshots, Thin Provisioning, Deduplication, Encryption

LINSTOR leverages on LVM and brings from that thin provisioning and efficient snapshots. It can add VDO into the storage stack to enable in-line deduplication. Via LUKS/dm-crypt it can encrypt data at rest. In all these areas it uses the existing building blocks on the Linux platform. Building on the shoulders of Open Source giants avoiding re-inventing the wheel.

NVME Over Fabric

LINSTOR is a very flexible system, with DRBD it is suitable for use with local storage devices, where compute nodes and storage nodes are hyperconverged. It can also use NVMe-oF to deliver storage from dedicated storage nodes to compute nodes or even from specialized NVMe-oF target devices like (WDs OpenFlex E3000). Thus orchestrate composable infrastructure / a disaggregated architecture.