I have made an interesting observation in my setup, and was hoping someone could take a look at these results and tell me if there’s an issue.
I have an Epyc 3251 server, 128GB RAM running XCP-ng. Using storage on Synology 1821+ w/ 64GB RAM, via 10GbaseT network via direct connection (dedicated port on server and dedicated port on NAS) using NFS.
Synology has both HDD based (RAID 6, 14TBx6) or SSD based (RAID 1, 4TB SATAx2) NFS storage, yet using Crystal Disk Benchmark the performance seems to be less than I would expect and also not at all different between SSD and HDD-based storage which was surprising to me. What was more suprising is that the performance was higher via SMB but also not different between SSD and HDD.
XCP-ng via NFS
SMB Via workstation
Does this look normal?
High density storage drives in raid can perform really well because the reads and writes are distributed across more drives giving you a better parallel performance. Worth noting that Synology in the past has performed slightly better with iSCSI vs NFS, but iSCSI is thick provisioned with XCP-NG so I generally don’t use it.
I am somewhat disappointed, because it seems like it obviates the point of the SSD array.
I imagine a large SSD-only array would be better, but that’s very expensive.
I recently built a 6x8TB SSD array using the Samsung QVO drives for a friend using the same Synology NAS, and it comfortably does 1GB/s flat reads and writes over the network, really impressive.
To me it seems like I should repurpose the 2-drive SATA array as it doesn’t seem to be doing anything at quite a high $/GB.