Since 2024, LINBIT® has partnered with the open source-focused enterprise data storage, compute, and virtualization solutions company, 45Drives. This partnership helps 45Drives bring a high-performing, low-resource consuming, block storage-based 2-node high availability (HA) solution to their customers. Their HA solution based on DRBD® is a middle ground between their single-node ZFS offering and the resource demands of a large-scale Ceph cluster.
This article covers the LINBIT partnership with 45Drives and highlights some exciting demonstration, performance testing, and setup work that the 45Drives team has done with DRBD running on some of the very potent hardware that they build. The results of the 45Drives team’s 2-node HA solution with DRBD is worth exploring, before discounting what a 2-node cluster can deliver, or before running to the hills for fear of creating a split-brain factory.
Background on 45Drives and LINBIT
45Drives is a North America-based ISO 9001-certified company that builds and ships complete hardware solutions with necessary software installed and preconfigured. The 45Drives support team of engineers can also work with its customers to help get systems up and running in a way that is optimal for particular environments or use cases. Systems that the 45Drives team designs and builds can focus on compute, storage, virtualization, or hyperconverged solutions.
LINBIT has partnered with 45Drives since a research project in 2024, when the 45Drives team was exploring options for integrating a pure block storage replication technology with VMware ESXi, Hyper-V, and other platforms. This was an identified need that their existing object storage-based data replication solution did not fulfill.
The research project went well and LINBIT and 45Drives formalized a partnership. 45Drives now offers its customers a DRBD-based HA clustering solution option.
In the 45Drives team’s own words:
45Drives DRBD clusters combine the best of both our ZFS servers and Ceph Clusters solutions. They deliver single-client transfer speeds comparable to ZFS servers, while providing high availability and server-level redundancy through real-time data replication between two servers.
Pacemaker and DRBD 2-node clustering in 45Drives setups
45Drives offers 2-node storage or hyperconverged system solutions that use Pacemaker and DRBD. The mention of a 2-node cluster should make systems administrators immediately think of the dangers of data divergence and split-brain scenarios. However, “2-node” does not have to be a dirty term. When an administrator properly configures fencing and STONITH, a 2-node Pacemaker and DRBD cluster can be a production-ready and dependable solution, just as many running-for-years deployments that the LINBIT team has supported worldwide can attest to.
For a setting up overview and demonstration on real-world 45Drives Stornado F16 all-flash hardware, 45Drives Chief Architect, Mitch Hall, made a great walkthrough video: “100% Uptime for Less: Building a 2-Node DRBD Highly Available Cluster with Pacemaker/Corosync”. In the video, Hall uses DRBD-backed ZFS Zpools as an example use case, and shows using Ansible to quickly and easily deploy DRBD in the test environment (with only a minor amount of monitor shaking involved). Hall also shows verification steps and failover testing. The video is a comprehensive introduction and well-worth your time.
Highly available virtualization solutions
Beyond compute and storage solutions, 45Drives also offers virtualization-focused solutions, by using Proxmox VE as the virtualization management platform. Proxmox VE is an open source alternative to VMware and Hyper-V. Proxmox VE users can integrate DRBD block device replication by using the LINBIT-developed, open source LINSTOR® Proxmox plugin. The LINBIT blog article, “Production-ready High Availability for Proxmox VE with LINSTOR” covers the LINSTOR and DRBD integration in Proxmox in more detail.
By using a clustered Pacemaker and DRBD-based solution, portable virtual machines and containers hosted on ESXi, Hyper-V, Proxmox VE, and other hypervisors or virtualization platforms can achieve near hardware storage performance similar to what is possible by using more expensive SAN or NAS appliances, without sacrificing resilience. The low latency access to storage that you get when using DRBD to provide block storage to your hypervisors makes hosting databases, message queues, and other highly transactional applications in VMs less of a concern in small cluster environments, compared to other solutions that use object, or file based storage.
When your hypervisor is Linux based, you can deploy a hyperconverged infrastructure (HCI) model where direct access to DRBD-replicated block storage is possible. If you are not using an HCI model, you can export DRBD block storage devices that might back your VM disk images by using standard block storage transports such as iSCSI. If your hypervisor supports the NVMe-oF implementation found in the Linux kernel, you can have even better storage performance by exporting NVMe-oF targets backed by LINSTOR volumes. ESXi, unfortunately, does not support the Linux NVMe-oF specification but Linux based hypervisors such as the 45Drives-supported Proxmox VE offering certainly do.
Benchmarking DRBD performance in a 2-node 45Drives environment
Another 45Drives-produced video with Mitch Hall, “DRBD Performance Test: Storage Performance on Stornado F16 All-Flash Servers with LINBIT”, dives into some interesting performance testing. In the video, Hall uses fio to do some before and after benchmarking. I won’t spoil the excitement of the results in this article but watching the video and seeing the numbers should help you appreciate one of the reasons 45Drives chose to partner with LINBIT and offer a DRBD-based solution to 45Drives customers.
In the two 45Drives videos mentioned in this article along the way, viewers can also learn about some feature-based benefits that DRBD can offer in 45Drives-built hardware. One of these is DRBD “diskless” client access mode which allows a DRBD peer node without local storage to access DRBD replicated storage from peers with local storage, over the network. Again, I won’t spoil watching the video in this article. Hall does an excellent job of presenting the utility of the feature.
Another DRBD feature that Hall touches on in the video is DRBD quorum. This feature applies to 3-or-more-node clusters. With the DRBD quorum feature, an administrator can use DRBD itself to avoid split-brain scenarios.
And, for technically-minded readers and viewers interested in a thorough explanation of how DRBD works “under the hood”, LINBIT Solutions Architect, Matt Kereczman, goes into much detail in the video.
Conclusion
The 45Drives team’s testing and benchmark results show that a 2-node Pacemaker and DRBD cluster need not be feared and can deliver high availability and near-hardware storage performance at a scale that is easier to manage for smaller teams.
For workloads that have outgrown a single-node ZFS server but do not yet need the complexity and resources of a Ceph cluster to achieve performance or large-scale needs, the 45Drives DRBD-based solution is a sound choice. The Stornado F16 benchmark results support the performance case, and an Ansible-based deployment keeps the setup process manageable and consistent. This is great for an enterprise that might need to deploy the same 2-node setup across many sites.
Whether your target workload is raw block storage, Proxmox VE virtualization, or storage exported over iSCSI or NVMe-oF to external hypervisors, the components are available and have been used in production. If you are evaluating HA storage options, the 45Drives setup and benchmarking videos are worth your time to watch. If you want to learn more about whether a LINBIT and 45Drives solution fits your environment, both teams are available. Reach out to 45Drives or reach out to LINBIT.
