Is it possible to use NFS as the transfer protocol for Hyper Backup from NAS #1 to NAS #2?

[jeyare]

I received this question over last weekend, when I was out of my digital kingdom
Here is a similar, but not same tested case for SynoDrive Team folder source based on mounted NFS Share from remote NAS. Works!

Back to this thread question

Testing scenario environment:
NAS#1 - data source NAS, Shared folder with NFS permission (v.3 is enough)
NAS#2 - backup target, Shared folder with NFS permission (v.3 is enough)

Basic:
1. You can create NFS Mount in File Station very easy, incl. useful parameter: Auto-mount after each DSM restart. Done. Easy
2. HB can create a backup task to Local folder.
3. Select the Backup destination - Local
4. Create Backup task - Shared Folder. Here is final stage of the test, because you can't select different folder than Shared folder. Then Mounted NFS isn't visible as choice for your Backup Destination. Then you can't define such NFS Destination for your Backup task.

Result: Busted

 

[akahan]

This is disappointing, of course, because NFS is the faster way to transfer. But now I am wondering if it really matters: When doing a Hyperbackup backup to another Diskstation on the LAN, what is the bottleneck? Is it transfer speed, computational power of the client, disk access speed of the client, or something else? ("Where is the bottleneck" may be different if we are compressing or encrypting the data, as well, as this requires more computational power from the client.) Does anyone know what is the choke point for these backups?

 

[Coop777]

your bottlenecks for transferring data between NAS devices are disk types/speeds, array configuration, and of course your wired Network speed. I have a share mounted from one to another and my transfer speeds from SHR-1 spindle array to single volumes on the other NAS are typically ~85 - 90 Mb/s on a 1 GBE network.

 

[akahan]

Understood of course that all of those things can be bottlenecks for transfers, just as the best speed a car can take around the track will depend on tires, engine, driver, road surface, etc.... The question here, though, is specifically related to Hyperbackup, which requires significant computational effort from the client.

So if we assume SHR1, 7200rpm NAS drives, 1GBE network, single volume on each NAS, where is the bang for the buck in improving backup speed? Add RAM? Add SSD cache? NFS (if that were possible)? 10GBE adaptors? Or is the processor the bottleneck, such that nothing else will matter?

 

[Coop777]

My 2nd NAS doesn't support 10 GBE, but I can say that Spindle drive on PC to NAS with 10 GBE will give you roughly 2x - 3x faster transfer speeds. So I would think that upgrading the network assuming both NAS devices support it would give you roughly the same.

 

[jeyare]

the problem is in deeper layer than HDD rpm parameter or 1Gbps network parameter.

1. Disk data saturation. Performance of every single PMR disk is degrading when they reach almost full capacity [frequently from >50%). Reason is simple - cylinders and structure of data write/operation. It’s HDD feature.
Solution, as I wrote many times - be precise in storage design calculations, also for 25% of “empty” space, which will save your performance. Keep your disk under 75%.

2. Disk geometry vs filesystem block structure vs network frames setup

3. Structure of data for the backup - lot of small files vs few big files
...