freenas + veeam/iscsi

[razvanc.mobile]

Hello everyone.

I plan on building another freenas server for my veeam backup needs.
i've been using a freenas setup for more than a year, and it's been fine, haven't had any serious issues with it.
But now, since it'll be a new thing, i've started digging a little more and i would like to do it right from the start.

Hardware-wise:
Dell R510 (12x3.5 chassis)
Dual L5630
32 GB ram
Will start with a raidz2 pool of 6 x 3TB WD Reds

Usage: Veeam B&R running on windows.
I've tried connecting from the veeam machine to freenas using cifs/nfs/iscsi, and i settled on iscsi (due to initial performance tests-basically the nfs implementation in windows has abysmal performance, cifs is somewhat decent, while iscsi is wirespeed).
This freenas server will not be used for anything else but those backups, which will be made using "Incremental" /w syntethic fulls. with 3 restore points on disks.
That means the the iscsi disk will contain 1 big file (100-500GB) + and 3 smaller files (<5GB) for each job, and there are currently about 20 jobs.

Now, that's the background(also worthy of mentioning is i'm using 1 gbps network).

Apparently there's an issue with high fragmentation for this type of scenario (currently looking at 33% FRAG on a similar server), and also i was planning on making one big iscsi disk that will probably fill all vdev, making freenas complain about pool usage from the start.(although currently i will be using less than 30% of the ~12 TB i will be getting from 6x3TB raidz2).

Given this pattern usage, any recommendations for the new setup?

Thank you for your time.

 

[diehard]

Don't make the zvol fill the entire pool. Only make it as big as you need it and let the rest be for free space, unused space in Windows is not the same as unused space on the pool.

Is your Veeam repository on a virtual or physical box? This can change how you send UNMAP commands. Use per-VM backup files and multiple IO streams. Why so many jobs? dedup only works with VM's within the same job.

 

[razvanc.mobile]

veeam is installed on a virtual machine, and the backup repository is the iscsi device from freenas. i can switch to a physical machine to run veeam if it would make a difference.

the space needed for backups grows daily, and that's why i wanted to make the zvol as big as the vdev from the start. i would gladly use a filebased solution instead of iscsi, but the performance difference was huge (think 10MB/s nfs vs 50MB/s cifs vs 110MB/s iscsi.) Maybe i should try again and compare, this was tested a long time ago , things might have improved?

Another way i could connect freenas to the veeam machine would be to keep veeam virtualized, and mount a nfs datastore in vmware, and create the backup repository on that datastore. would that be a better solution?

as stated earlier, current solution works ok (virtual veeam with backup repository as an iscsi drive from freenas), but lately i see scrub times much higher than before (70 hours) but my current pool (2 vdevs, each made of 4x2TB raidz1) is 33% fragmented and 83% full.

Sent from my SM-G930F using Tapatalk

 

[Mlovelace]

Why not do a tiered approach since space is limited. Do a fast reverse incremental to the iSCSI zvol so your latest is always a full, and keep a couple days on that. Don't fill more then 50% of the LUN with the RI job. Then do a backup copy job to a CIFS repo (use a proxy to keep the IO out of production) to let you store more fulls. Don't fill the pool past 80% or performance will tank.