Troubleshooting Slow Speeds on Proxmox Backup Server (PBS)

- Proxmox VE host

- TrueNAS server (10Gb link)

- PBS running as a VM inside TrueNAS

- Storage: 4 HDD RAIDZ1 pool

- Observed speed: ~250 MB/s (~2 Gbps)

250 MB/s equals roughly 2 Gbps.

On spinning disks in RAIDZ1, that may already be close to the practical throughput limit, depending on block size and workload.

The PBS virtual machine had too much RAM allocated, which left too little memory for the ZFS ARC cache on the TrueNAS host.

When ZFS does not have enough ARC cache available, disk I/O performance drops significantly, and backup workloads suffer most.

- ZFS relies heavily on RAM for ARC caching

- PBS also uses memory for deduplication and indexing

- Allocating too much RAM to the VM starves the host

- Less ARC means more disk hits and slower backups

- Verify 10Gb throughput using iperf

- Check TrueNAS ARC usage during backup

- Monitor CPU utilization (compression and dedup)

- Confirm RAIDZ1 pool performance baseline

- Adjust VM RAM allocation and retest

- Leave enough RAM for ZFS ARC (the host comes first)

- Avoid over-allocating memory to the PBS VM

- Consider SSD/NVMe for the PBS datastore if you need higher speeds

- Use proper block size alignment

Should 10Gb always reach 1.2 GB/s?

Only if storage and CPU can sustain that throughput. A fast network does not guarantee disk performance.

Does RAIDZ1 limit throughput?

Yes. RAIDZ1 on spinning disks often performs closer to single-disk speeds for certain workloads.

Is PBS in a VM a bottleneck?

It can be if CPU, RAM, or storage resources are constrained.

What was the final fix?

Reducing the RAM allocated to the PBS VM so ZFS ARC had enough memory.