Pool & vdev design for backup Server

mangoon

Dabbler
Joined
Dec 18, 2022
Messages
13
Hey guys,

I really need your help deciding on planning my Pool for my Backup Server...

I mainly run a QNAP 8x8 TB RAID6 on Ironwolfs and until now, I backed up all important stuff to an old QNAP with 4x4TB RAID 0, that just went only once a day, made the copy and then shutdown. So far, that was a pretty good solution, as I also have a remote Setup that is practically the same but running 24/7.

Now in the new house and finaly with a full equiped 19" rack, I decided to slowly replace all the stuff and start with the "offline Backup" machine. I got a 2U Supermicro Chassis with an E3-1231v3 on a Supermicro X10SLL-F with 32gb of ECC ram and 2x LSI 9211 in ITmode.

Also I got some of my older HDDs and two used SAS drives from ebay, that I have now 14x 4TB HDDs (mixed in all varieties) and am planning my pool Layout as I wait on the last parts of the build to arrive.

I decided to go with TrueNAS Core for now as I only just need it as a backup target for my main NAS and desktop PC.

I originally wanted to go 12x4TB in RAIDZ 2, having the 2 extra disks just laying in the room if one drive fails, but am concerned now a little about performance and resilver & scrub times... Is it better to just go with 2x6 drive RAIDZ1 in two vdevs? I mean I only have gigabit for now, not planning to go 10G in the near future.... Would the single RAIDZ2 pool have enough performance to deliver constant write speed of ~ 200MB/s? (I plan on LACPing and backing up NAS and desktop at the same time).

How about resilvering and scrubbing? Will the machine do that in a resonable amount of time? And do I really have performance benefits from 2x RAIDZ1 compared to 1x RAIDZ2?

Also, when I go the 2x RAIDZ1 train, should I connect eache vdev to one HBA? Or mix?

And a side note: The redundancy is not that important as its just my backup machine, so if I loose the pool, I will just build a new one and do the backup again. That WILL be a pain in the butt, but I'm not too concerned about that...

Questions over questions... There I'm happy to know, that my offsite Backup with 4x6TB SAS in a microserver Gen8 ist just plain simple to build and decide :-D
 

mangoon

Dabbler
Joined
Dec 18, 2022
Messages
13
A another follow up Question: If I decide to go with 2x6 drive RaidZ1, I understood, that I can just start with 6drives, copy the initial data and then creat the second vdev inside the pool? How about balaning the space? If I lets say copied alrady 10TB of data on the first vdev, what happens after adding the second? Does it just use the second vdev until its balanced out?

Additionally, When I decide to replace my main NAS and use the drives from there for the backup server, I understood, that I coudl just switch the drives one by one in on vdev until I got all six replaced? How about pool balancing then, when I have 6x8TB and 6x4TB?
 

Davvo

MVP
Joined
Jul 12, 2022
Messages
3,222
Welcome!
Read the following resource to gain a basic understanding of how ZFS works.

Also I got some of my older HDDs and two used SAS drives from ebay, that I have now 14x 4TB HDDs (mixed in all varieties) and am planning my pool Layout as I wait on the last parts of the build to arrive.
Oh the pain. Make sure they are all CMR. Also do note if there are differences in RPM. Make sure to properly test your drives, especially the used ones. Read the following resources.

With such a number of drives you want a proper PSU, please read the following resource.

I originally wanted to go 12x4TB in RAIDZ 2, having the 2 extra disks just laying in the room if one drive fails, but am concerned now a little about performance and resilver & scrub times... Is it better to just go with 2x6 drive RAIDZ1 in two vdevs? I mean I only have gigabit for now, not planning to go 10G in the near future.... Would the single RAIDZ2 pool have enough performance to deliver constant write speed of ~ 200MB/s? (I plan on LACPing and backing up NAS and desktop at the same time)
Since you plan on using it for backup (with very few simultaneous users, I think that a single vdev with all 12 disks in RAIDZ3 is your best option (altough you would be at the limits of a vdev max suggested width).
Please read the following resource in order to predict your pool performance.

For LACP you can read this other resource.

How about resilvering and scrubbing? Will the machine do that in a resonable amount of time? And do I really have performance benefits from 2x RAIDZ1 compared to 1x RAIDZ2?
What do you consider reasonable?
If you have read the resource about pool layout I linked you should already have the answer: main performance benefit is the increased IOPS (that your use case doesn't need) at the cost of significant risk. Consider RAIDZ1 dead alongside RAID5.

Also, when I go the 2x RAIDZ1 train, should I connect eache vdev to one HBA? Or mix?
It doesn't matter, a proper HBA in IT mode will act like an extension of your motherboard, virtually increasing the SATA ports on it. ZFS will recognize each device.

And a side note: The redundancy is not that important as its just my backup machine, so if I loose the pool, I will just build a new one and do the backup again. That WILL be a pain in the butt, but I'm not too concerned about that...
Then ZFS (and as such TrueNAS) might not be the right solution for you: ZFS focuses on data integrity, nothing comes before it. You can see it as "all data or no data".

A another follow up Question: If I decide to go with 2x6 drive RaidZ1, I understood, that I can just start with 6drives, copy the initial data and then creat the second vdev inside the pool? How about balaning the space? If I lets say copied alrady 10TB of data on the first vdev, what happens after adding the second? Does it just use the second vdev until its balanced out?
Regarding rebalancing I am not sure: iirc yes, the new data should be written in the emptier vdev, but there should also be a few community made scripts available.

Additionally, When I decide to replace my main NAS and use the drives from there for the backup server, I understood, that I coudl just switch the drives one by one in on vdev until I got all six replaced?
Yes, you can switch the drives the way you wrote (offline, resilver, repeat). There are a few tricks that can be used (like replacing more than 1 disk simultaneously).

How about pool balancing then, when I have 6x8TB and 6x4TB?
It is reccomended to have symmetrical vdevs.

Have fun reading and forgive me for any spelling issue. It was tough writing all this from a smartphone.
 
Last edited:

mangoon

Dabbler
Joined
Dec 18, 2022
Messages
13
Thank you so much for taking that much time, I will repay this in answering your comments as good as possible. Just one little thing to myself: I work in IT now for over 15 years, run my own homelab since the beginning and am quite qware about some querks... But I made my journey from selfbuild Windows Servers for years and jumped the QNAP Train some time ago. But I'm annoyed beyond a lot of things about how QNAP handles alot of stuff and am looking for a replacement for my whole setup. After a long journey in reading about different new ways of managing my storage AND now finaly the possibility to move all my stuff to a rack where noise and space is nearly no concern I'm ready to beginn a new chapter and give something new a try. And after fiddling around with OMV, Synology DSM (on custom hw, cause the synologys have sooo weak hardware, windows Server (again) I finally fell in love with how truenas handles everything and am quite sure that I really want to invest time and money in revamping my whole setup. And for that I want to beginn with the least critical bit of hardware and thats the offline backup System.

Also to understand my situation more, here is an overview of everything I run atm and what is already planned:

Main NAS: Qnap 873 with 8x8TB Ironwolf RAID 6 - Acutally loaded with ~19TB of stuff, where I have ~12TB of really important data (such as photos, documents etc. I'm a part time photographer and drone pilot and really need those backed up properbly)

Online Offsite Backup: Qnap 431 with 4x4TB WD Green RAID0 - It just runs... I had to switch from resilio to RTRR due to lack of memory, which is kind of a pain in the ass, as the buddy of mine, where it is located sometimes messes around with his router, so I plan to replace that real soon, also cause it gets a little full. Just replicates the "Important stuff"

Offline Onsite Backup: Qnap 431 with 4x4TB Seagate Desktop RAID0 - That is the one that I want to replace first and what is the thread all about. It boots up every day at 2 PM, pulls the important stuff and shuts down afterwards. That is just to safe some power and protect me against any crypto virus and stuff.

Cloud Backup: I have all my photos on amazon and everything else, that is important on onedrive for business, except videos which get more and more and I'm not sure atm. how to handle those...

So basically I'm fine, running even RAID0 on my backups, as I only had one drive dying over the last years and it was easy to replace.... I had to resync the whole machine, but that was the price I was happily willing to pay for saving some money on drives.

What I have generally planned in 2023:
Build the new offline Backup Server (thats what the post is about) and backup the whole NAS instead of just the important bits. Its supposed to handle the whole data thats on it right now and also have enought free space to load some crap off if I need, lets say a second space for something temporary, like redoing a friends PC etc. I also try to figure out, how I want to sync everything and if I can shut it down when not needed and how truenas handles being only runnign a couple of hours a day. And yes, I know, that a lot of NAS drives are meant to run rather than start, but I dont want to have 12 drives spinning 24/7 when I only need them 30 minutes a day. And if i let them spindown, there is really now reason in my opinion to nut just shutdown the entire system.... I think I will stick with running it once a day and then lets say on sunday keep it up for the whole day for scrubs etc.

Replace the Online Backup: As soon as the offline Server is done, I have a Microserver Gen8 sitting next to me with 12 gb ECC RAM and a G1610 CPU. I wait for 4x6TB SAS drives in and a HBA I have ordered for it to replace the old junk... I will put it in RAIDZ1 at first, but am also pretty confident in running it as a stripe, if it gets too full. I don't have any performance concerns for that one, cause it will only get its data via my 40mbit upload, so that should be fine...

PC Backup on demand: I currently backup all my OS disks from my Desktop and all our notebooks with Veeam and it just works. But: I have together 8TB of SSD storage in my desktop and, as there is nothing really important on it, as everything important is also on the QNAP or stuff like steam games etc. I want to be able to backup everything, so that I don't have to redownload everything and so on. For that I have a Bluechip Server with a 1245v5 and 16GB ram sitting here, together with 5x3TB drives, that I still have laying around and will be just a stripe (or Z1, still deciding) and will be booted up with WoL once a week, get synced and then remote shutdown. I also want to use that one as a playing device for trueNAS as I don't want to touch my productive machines with new releases etc.

And when I'm done with all that, I plan on getting a new 12 (or16) bay server end of next year to replace the QNAP. I plan on reusing the 8TB Ironwolfs for backups and will then see, If I really get going with truenas for my production too or go a different path.... But thats something for next summer...

So maybe some of my questions and answers below will be in better context now, as I (at least) think I know pretty much, what I'm getting into with but am always open to suggestions


Welcome!
Read the following resource to gain a basic understanding of how ZFS works.
I did that before the Post, but thanks again for the links, its handy to have them all in one thread for further reference :-D

Oh the pain. Make sure they are all CMR. Also do note if there are differences in RPM. Make sure to properly test your drives, especially the used ones. Read the following resources.
I read about it and will check soon if there are any SMR drives in my collection and will see.... I really don't think that SMR will be that much of a hassle in my particular case but want to avoid it if possible. Maybe, if I have more than two SMRs I will grab another pair of used 4TB SAS drives from ebay, I only paid 50€ for the last two I ordered, so that is still to be decided.

With such a number of drives you want a proper PSU, please read the following resource.
Yeah I know... My chassis has a 560watts PSU that came with the case... I will use it for now and monitor the power draw with an external monitor and then maybe replace it. Also all my stuff is backed with an online UPS, so at least I know that the power it gets ist "good" :-D

Since you plan on using it for backup (with very few simultaneous users, I think that a single vdev with all 12 disks in RAIDZ3 is your best option (altough you would be at the limits of a vdev max suggested width).
Please read the following resource in order to predict your pool performance.
Thats what I figured. I mean, atm I only want to sync it with the main NAS via 1gb LAN, so that should not be a problem at all. I mainly thought about doing two RAIDZ1 vdevs for other stuff. If I could grow the vdevs (like in QNAP) i would not even think about not using RAIDZ2 and just start with less disks and expand later. Or even go with RAIDZ3 (having in mind how old some of the drives already are and be done with it...

For LACP you can read this other resource.
Yeah I know... I mean my QNAP runs on LACP, so that if my backup runs the same time as someone wants to use plex there is enough bandwith but I don't know if I don't just stick on not using it for the moment... It was just a thought that maybe later, its worth enabling it. I also think about upgrading to 10g but that would cost quite a lot as I would need a switch and a couple of cards... So I will see in the future.

What do you consider reasonable?
If you have read the resource about pool layout I linked you should already have the answer: main performance benefit is the increased IOPS (that your use case doesn't need) at the cost of significant risk. Consider RAIDZ1 dead alongside RAID5.

It doesn't matter, a proper HBA in IT mode will act like an extension of your motherboard, virtually increasing the SATA ports on it. ZFS will recognize each device.
Yeah, thats why I asked in the first place... I don't mind the risk (I run it with RAID0 atm), but was hoping to sicnificantly reduce srubbing and resilvering times with two smaller pools. I really don't want to use more than two disks for parity all together as I want to be able to use the thing quite a little time without it getting full...

Then ZFS (and as such TrueNAS) might not be the right solution for you: ZFS focuses on data integrity, nothing comes before it. You can see it as "all data or no data".
Yeah I know and thats what brought me here in the first place as I really want that for my main NAS, as soon as I am ready to replace the QNAP and am familiar with truenas enough to make the full switch. As stated above, thats also the reason I want to start with the "unimportant" machines, rather then fully switch right now

Regarding rebalancing I am not sure: iirc yes, the new data should be written in the emptier vdev, but there should also be a few community made scripts available.

Yes, you can switch the drives the way you wrote (offline, resilver, repeat). There are a few tricks that can be used (like replacing more than 1 disk simultaneously).

It is reccomended to have symmetrical vdevs.

Have fun reading and forgive me for any spelling issue. It was tough writing all this from a smartphone.
Does anyone else have more insights here? Do I get any benefits in having two smaller RAIDZ1 vdev rather than one big RAIDZ2 vdev other than IO performance?
 

ChrisRJ

Wizard
Joined
Oct 23, 2020
Messages
1,919
I did that before the Post, but thanks again for the links, its handy to have them all in one thread for further reference :-D
In that case I would strongly recommend to read things again, because a lot of your questions are actually covered there.

I read about it and will check soon if there are any SMR drives in my collection and will see.... I really don't think that SMR will be that much of a hassle in my particular case but want to avoid it if possible.
Well, you are not the first to think this. As they saying goes: It's your funeral. Seriously, you should not brush off the warnings like this. SMR drives absolutely don't work well with ZFS.

Yeah I know... My chassis has a 560watts PSU that came with the case... I will use it for now and monitor the power draw with an external monitor and then maybe replace it.
Well, if you think that a power monitor (unless we are talking about professional and calibrated gear for a couple of thousand bucks) is able to measure the peak current from HDDs spinning up, have fun.
 

Davvo

MVP
Joined
Jul 12, 2022
Messages
3,222
There is a reason I linked you to resources, please take your time to read them.

Does anyone else have more insights here? Do I get any benefits in having two smaller RAIDZ1 vdev rather than one big RAIDZ2 vdev other than IO performance?
Most, if not all of your questions regarding pool composition and performance can gain an answer by reading the ZFS Storage Layout resource that I previously linked.
Screenshot_2.png
This is the pool failure curve of 3 different configurations:
The blue one is the single vdev 12 wide RAIDZ3;
The yellowish green one is the single vdev 12 wide RAIDZ2;
The red one is the 2 vdev 6 wide RAIDZ1.
The shallower the curve is, the lower the probability of losing your pool is when you lose a drive.
Compared to the other two, the red one is a disaster.
The usable space (80% of pool occupancy) you have in each config (with 4TB drives) is: 26.95 TB, 28.24 TB and 29.65 TB.
The red option (2x6 vdevs in RAIDZ1) is not worth imho, and adding another parity disk to each vdev (2x6 vdevs in RAIDZ2) would bring the space down to 24.71 TB. For reference, I used this calulator.
Going single vdev is best from reliability and capacity and since the use if for (secondary) backup you don't need stellar performance.

Yeah, thats why I asked in the first place... I don't mind the risk (I run it with RAID0 atm), but was hoping to sicnificantly reduce srubbing and resilvering times with two smaller pools. I really don't want to use more than two disks for parity all together as I want to be able to use the thing quite a little time without it getting full...
The bigger the drives' size, the longer the resilver time. You still haven't answered my question: please quantify what you consider reasonable time. Also, why do you care so much about performance since it's a (secondary) backup server?

Yeah I know... I mean my QNAP runs on LACP, so that if my backup runs the same time as someone wants to use plex there is enough bandwith but I don't know if I don't just stick on not using it for the moment... It was just a thought that maybe later, its worth enabling it. I also think about upgrading to 10g but that would cost quite a lot as I would need a switch and a couple of cards... So I will see in the future.
We have a nice resource if you are interested in 10 Gbps.

Thats what I figured. I mean, atm I only want to sync it with the main NAS via 1gb LAN, so that should not be a problem at all. I mainly thought about doing two RAIDZ1 vdevs for other stuff. If I could grow the vdevs (like in QNAP) i would not even think about not using RAIDZ2 and just start with less disks and expand later. Or even go with RAIDZ3 (having in mind how old some of the drives already are and be done with it...
Right now the only way to change a RAIDZ vdev is to destroy the pool an recreate it. You want to accurately plan your pool structure since it's not possible to add/remove drives to single vdev, as well as removing a vdev. The only thing that's possible is to add another vdev.
RAIDZ1 is strongly discouraged (raid5 is dead) with 3TB or larger drives since losing another drive during resilver (due to it being a stressful process) means total loss of the pool.
In ZFS (TrueNAS) you really need to plan your system and try to predict what will be your future needs.

Finally, I do think that you could reduce the number of systems that you use since you said that you currently use 3. You could just use one onsite and one offsite alongside your cloud storage (that too could be replaced, but it's a different topic).

Edits were made for spelling corrections.
 
Last edited:

mangoon

Dabbler
Joined
Dec 18, 2022
Messages
13
Thanks a lot for all the replies. I think I'll just go with a 12 wide Raidz2

What I meant with resilber times is the following: Lets say, I hav 20TB of data,thats on there. Then just recopying it would rougly take 55-60 hour with 1gbe, IF it can saturate that. Now if resilvering, after a drive failure would need significantly longer (lets say 2 weeks), I would rather rebuild the pool with a new drive and copy everything again, given its just the backup machine... So thats why I ask... I personally am also totally fine with a dumb stripe if resilvering really needs long long long.... But you always only get "Raodz2 takes longer as its double the parity. The larger the vdev the longer it gets. But what" longer" means, noone ever tells In a classical RAID6 its easy, there you just write the whole drive new on a blocklevel and if you easily get 100mb/s for transfers it also gets easily 100mb/s for rebuild. So you can just calculate how long it should take...

And why I'm concerned about performance: I want at least to saturate 1gbe (what I will surely do, given the fact, that every RAIDZ (1-3) will excell in sequential writes) but I dont want to miss an portant "boost" here designing it now wrong and later hating myself...
 

Davvo

MVP
Joined
Jul 12, 2022
Messages
3,222
I believe you can expect a resilver time under 48 hours.
Depending on the files and how you configure your datasets you should saturate 1Gbps.
 
Top