Enhancement of NFS repository
NFS repository is rely on NFS protocol, which is not built for heavy loads and using it should be used only, if another setup of repository is not possible.
In my usecase is the only option, how to avoid connection of guests to repositories.
NFS repository does not support fast cloning, so consumption is much higher like on repository with direct attached storage or object storage.
Synthetic backup is taking very, very long for big VMs especially.
How synthetic backup works with NFS repository
During process gateway server needs to take files on gateway, where is data mover, create synthetic full backup and send it backup on nfs storage.
How NFS repository works: they are mounting during backup/restore process in one threat and unmounted after backup/restore is done:
Such configuration is very safe and is not visible nfs export during non-runtime.
Based on my test one threat can run backups around 21MBps on 1Gbps line.
In this case was distance was 50km and some connection was over internet as well.
To increase speed of backup is needed to increase threats.
So we have been mounting nfs export permanently with 16 threats/connections, where was increased speed to 96. More threats did not speed up anymore. So 16 threats is optimal.
| Nconnect | Speed |
| 4 | 36,8 MBps |
| 8 | 88,8 MBps |
| 16 | 96.3 MBps |
If you see as bottleneck with NFS repository, this is good way how to increase speed with repository with nconnect.
Nconnect is tied up with ip, so if you use more paths, is ok. It will be applied for all of them. So you need to permanently mount nfs export with nconnect and effect is applied :)
My test were done with NFS4.x as based my security NFS3 is not secure ….
