I have a virtual file server cluster of 2 VMs, where each VM is running with Windows Server 2019 (8 GB RAM, 8 vCPUs). The file servers store their data on a VHD set, which is formatted with NTFS and has deduplication enabled. There is about 14 TB of data, which only consumes around 6 TB due to deduplication.
I have a physical server (64 GB RAM, 2 CPUs à 8 cores with HT, so 32 logical cores in total) with Windows Server 2019 and Microsoft SC DPM 2019 installed. It uses local HDDs (RAID6) to store the backup data on Modern Backup Storage (MBS), formatted with ReFS. This server has a redundant 10 GbE link to the file servers in the same subnet. Protection groups, that back up entire VMs or SQL databases perform well. But the protection group that backs up the data of the file server cluster has a really bad performance.
A full synchronization of the 14 (or 6) TB takes around 70 hours! That's incredibly slow. When it starts, it has about 2-4 MBit/s throughput on the 10 GbE link (which has its full potential when I copy a large file manually). After a long phase of slowness (~1 day) it "speeds" up to 200-400 MBit/s which is still quite slow.
CPUs and RAM do not seem to be a bottleneck on any server.
I found similar problems here and elsewhere:
- DPM for File Server - awfully slow?
- MS DPM - Slow performance of consistency check
- Slow Backup Speed with DPM after few minutes backup is started
- DPM 2016 MBS Performance downward spiral
- How can we improve SCDPM?
- DPM 2016 File Server Backup SLOW
But without a working solution. How can I speed up the backup of data?