Our consulting engineering company stores its data on-premises. We currently have two production Ubuntu file servers and two replicas, using two ZFS RAIDZ2 pools of roughly 150 TB each, with about 40–60% of each pool occupied. We recently decided to use Veeam for backing up both our physical infrastructure and file servers, but I discovered that Veeam cannot perform agentless backups of ZFS. Jobs also fail immediately because of the dataset size and the 218 TiB VBK limit, although I am not sure exactly how Veeam calculates that limit.
Most of our other production systems run Windows Server, so I am considering replacing the file servers with Windows and using DFS Namespaces. Another possibility is Azure Files with hot/cold tiering. The shares contain large LiDAR scans, drone-based topographic reports, and ordinary Office documents.
Would it be better to migrate to Windows or cloud storage, redesign the Veeam jobs, or use a different backup solution while keeping the existing ZFS servers?
4 Answers
You do not necessarily need to replace the ZFS systems. Veeam will not provide an agentless ZFS backup, but you can use a file-share backup job over SMB or NFS, or install the Veeam Agent for Linux on the servers. At this scale, split the data into multiple jobs by department, top-level share, or recovery requirement instead of creating one enormous chain for an entire 150 TB pool.
ZFS snapshots and replication with send/receive can provide fast local recovery, while Veeam can protect the important datasets with an off-site or immutable copy. Keeping separate jobs also means that one failure does not affect every restore point. Document the recovery process and perform real restores, including a large project restore, before relying on the design.
Do not replace a functioning file-serving platform solely to make one backup product easier to configure. First define the recovery targets: how quickly a single folder must be restored, how much data changes daily, and how much history must survive a site failure. Then test representative LiDAR, drone imagery, and Office data with their real permissions and paths.
Object storage may be worth considering for the off-site copy, particularly with a dataset this large, but validate the costs, retrieval times, immutability, and restore workflow. The best choice is the one that meets documented restore-time and disaster-recovery requirements, whether the primary file server runs Linux or Windows.
It is also reasonable to compare backup platforms instead of rebuilding the file servers. Products aimed at large unstructured datasets may handle NAS data and object storage differently, and some can avoid creating one massive backup file. However, since most of the backup infrastructure is already deployed for the HCI environment, changing products may create operational complexity and migration costs. Compare the complete design—including storage, licenses, cloud egress, retention, and restores—rather than just the initial quote.
Windows file servers with DFS are a well-tested combination with Veeam, especially when the storage is presented as manageable virtual disks or SAN volumes. I would still avoid assuming that DFS alone solves the backup design. Test the expected volume sizes, ACLs, restore performance, and licensing requirements first. Veeam support may also be able to recommend a suitable layout for your environment.
Most of our backup infrastructure is already deployed for the HCI environment, and we are already using Veeam there. I am mainly trying to determine whether adapting the file-server backup design is preferable to replacing the storage platform.