Veeam and RAID Level

[JScott]

Hi all,

 

Our current Veeam server is using a RAID0 because of its age and speed. I am looking at replacing the server but I am wondering what RAID level you are using on your Veeam backup servers. A full VM backup is about 2TB + daily incrementals of about 100GB.

 

Any feedback on speed trade off against RAID level and Veeam would be appreciated.

 

Regards

Jim

[Steve21]

Veeam storage wise we're on 5 or 6, (can't remember which they set it to originally), as our main backups run evening/weekend don't really see the speed factor being an issue with it.

 

Or did you mean OS wise?

 

Steve

[JScott]

I was meaning the RAID set on the hardware/server level. I am looking at a new Dell server. RAID0 would give me 32TB backup storage. I think RAID5 gives me 24TB storage. But after reading some other forum posts they criticise RAID5 and Veeam together. Are you using SureBackup as well?

[Steve21]

On our veeam server we have the OS RAID, and then the backup RAID. BackUp is either 5 or 6, and then dedup'd too

 

We have it on yes Seems "iffy" to say the least though lol

 

Steve

[JScott]

Thanks for the replies Steve,

 

Our SureBackups seem ok at the minute. 'iffy'?

 

Regards

Jim

[DMcCoy]

32TB RAID 0 isn't a *backup* it's a gamble. Or a clean slate...

[featured_spectre]

RAID0 for a backup...you're extremely brave. Lol

[JScott]

The current RAID0 was setup by an external company, when getting a new server I want to move away from this for obvious reasons. I was planning to setup the new server with RAID5 for the storage and RAID1 for the OS.

 

Thanks for the replies.

Jim

[ijk]

RAID 5 (and possibly 6) are not currently recommended because of the time it would take to rebuild the array if one of the disks fails, and you have an enormous array.

 

Another consideration that you'll find in various forums is that disks purchased together may be likely to fail together. The worry being that while you are rebuilding the array with a new disk, another fails.

 

RAID 10, while expensive, is desirable because more than one disk can fail without losing the array.

 

At least that's roughly what I recall from the hours of reading I did prior to choosing RAID 10 for our Veeam storage.

 

I'd check the details of the above before committing but I'm pretty sure that's what you'll find with research. There is quite a lot about this on the spiceworks forums.

 

Nic

[kmount]

Hi,

 

This may help you weigh up your options - Understanding the Impact of IOPS on Backup Performance: Best Practices

 

RAID10 preferred for performance reasons, but be careful assuming you can lose multiple disks .... it depends which disks! ... if it happens to be the disks in the same group (ie both AA rather than an AB) you will have problems. There are tactics to mitigate against some of this but don't think it means you can lose half of your drives and still survive. (you could if it was a single A B C D etc but if it's an A and A it could spell disaster) - A fun explanation of this from this chap:

 

A good explanation of this - hardware - Which RAID solutions can handle 2 disk failures - Server Fault

 

The big challenge here is finding the balance between capacity and performance, both during backup, and also during recovery especially if you use instant recovery and want to offer reasonable performance during a disaster. SureBackup etc will also thank you for good performance.

 

RAID0! The external company has set you up for a fail and I'm glad you've got wind of it and are putting it right.

[PotNoodleTech]

I'm running my Veeam server with raid 5 at the moment even though it is not best practice, it gives the only mixture of performance and capacity that I could use really (only 4 drives).

 

I'll upgrading next year.

 

To be honest take what we all say here with a pinch of salt because when we get down to the nitty gritty, ABSOLOUTLY ANYTHING IS BETTER THAN RAID 0.

[kmount]

We have customers using RAID5 and RAID6 on repositories and it is fine, acceptance of the limitations is all that's needed.

[featured_spectre]

Where I worked previously, the RAID for the VEEAM was 0, 2 days later, I convinced the line manager it had to be changed, so we got RAID10 instituted. 3 months later, there was a drive failure, so they were thankful it got done so quickly.

[aleach2]

We have a QNAP NAS with RAID 5 setup as a CIFS (SMB) Share. This gives us 19TB of storage. We have our Veeam server as a virtual server backing up to the QNAP NAS and replicating to another NAS across the site. Works very well and since going virtual with our Veeam server (which I was sceptical about) the performance has increased.

 

Would not recommend RAID 0. Anything is better than that.

[Michael]

I have to agree with others comments - RAID0 for any backup array (irrelevant of size) really isn't a good idea.

 

I'm speculating your servers themselves are either RAID1, 5, 6 or possibly even 10, so it wouldn't be unreasonable to think your backup array should be the same or better.

 

Even better - you could argue with that volume of data, having two separate backup arrays would be the way to go.

 

I'd rather have 2 x 12TB, rather than 1 x 24TB for example. Again this does greatly help/improve your chances of not losing data.

[JScott]

Thanks for all the replies. I am now thinking of a RAID10 with 8x4TB HDD. Giving a 16TB total.

 

Next question. What is the best RAID block\stripe size to use when creating the array. The 16TB partition will be NTFS.

 

Any thoughts?

 

Regards

Jim

Edited May 20, 2015 by JScott
typo

[Jamo]

Thanks for all the replies. I am now thinking of a RAID10 with 8x4TB HDD. Giving a 16TB total.

 

Next question. What is the best RAID block\stripe size to use when creating the array. The 16TB partition will be NTFS.

 

Any thoughts?

 

Regards

Jim

 

I would say RAID6 to be honest, are you going to require RAID10? The performance gain is mostly on write, unless you are backing up gigantic data its not required. You would be better off being able to keep more restore points for longer than to write them to disk faster!

 

Regarding blocksize, NTFS and the newer filesystems don't tend to fragment so much due to the way they allocate the file blocks. The VEEAM files are also huge so block size isn't important. Partition alignment is, but I believe the OSes are dealing with that by default now.

 

https://support.microsoft.com/en-us/kb/140365

https://technet.microsoft.com/en-us/library/dd758814%28v=sql.100%29.aspx

[Davit2005]

Raid 5 with one spare here, 24 X 2TB disks this gives us about 40TB to play with. We have about 30TB of Veeam backups with Retention period set to 28days.

 

One thing I would suggest is to use a bigger file allocation size than you actually need. We originally had 12 X 2TB but since adding a needed extra shelf for capacity we have backups split between 2 volumes and will probably need a reformat to sort out . You can end up with a 1KB file taking 8K or 16K on disk but you probably won't be having a lot of small files anyway.

[PotNoodleTech]

If you're only backing up Veeam/Vmware/HyperV images - i.e. massive files, I'd use a 64k block size.

 

If you're backing up individual file level then stick with 4k probably. It only makes a sight difference to speed anyway from the testing I've done.

[JScott]

Thanks for all the replies on this subject. Space wise RAID 6 does give a good trade off between fault tolerance and speed.

 

I am planning to stick 2012R2 on the server as the OS that Veeam sits on. Anyone running Veeam with the main backup repository running deduplication?

 

Jim

[Jamo]

Thanks for all the replies on this subject. Space wise RAID 6 does give a good trade off between fault tolerance and speed.

 

I am planning to stick 2012R2 on the server as the OS that Veeam sits on. Anyone running Veeam with the main backup repository running deduplication?

 

Jim

 

I have had experience with that. It does work, however the 2012 dedup is not inline so we had to seriously up the scheduled tasks to actually achieve anything like useful de-duplication performance.

 

Space savings once it actually did dedup were impressive but were nearly wiped out during every full backup. Also you can't successfully run reverse incremental with it as it will rehydrate the whole backup each night meaning 2012 would have an insane amount of changes to dedup every backup cycle.

 

Personally I am much more a fan of arcserve's global dedup, its a huge improvement over veeams per job dedup and it means you no longer have to rely on the 2012 nightly optimisation jobs to fix the space issues.

[Steve21]

We run full dedup on our entire backup sets and don't have any issues using reverse incremental on all our backups. Yes we've got dedup running 24/7 on the Veeam server as it's not like it's doing anything else to need to have the CPU/RAM spare. (Guess that depends if you're using physical box etc)

@JScott Few suggestions:

 

If you are going to move it onto it make sure you have 0 days set for the backups to start deduping so you do get the dedup instantly

 

And get the jobs for dedupe-friendly compression, not optimal etc else it'll interfere with the Windows dedup and get worse results

 

Make sure you use more servers within a backup job as Veeam only dedupes within a job (on top of Windows dedup), lots of small jobs won't get best results.

 

Currently running at 10.8TB "saved" in dedup at 52% rate on our repository.

 

STeve

[Jamo]

We run full dedup on our entire backup sets and don't have any issues using reverse incremental on all our backups. Yes we've got dedup running 24/7 on the Veeam server as it's not like it's doing anything else to need to have the CPU/RAM spare. (Guess that depends if you're using physical box etc)

@JScott Few suggestions:

 

If you are going to move it onto it make sure you have 0 days set for the backups to start deduping so you do get the dedup instantly

 

And get the jobs for dedupe-friendly compression, not optimal etc else it'll interfere with the Windows dedup and get worse results

 

Make sure you use more servers within a backup job as Veeam only dedupes within a job (on top of Windows dedup), lots of small jobs won't get best results.

 

Currently running at 10.8TB "saved" in dedup at 52% rate on our repository.

 

STeve

 

Best Practice for MS Server 2012 DeDup Repo | view topic

 

This thread has some good info on the dedup and veeam. Tbh the issues I saw were likely due to the size of the backup files which many are way over 1TB.

 

https://msdn.microsoft.com/en-us/library/hh769303(v=vs.85).aspx

 

MS apparently doesn't support dedup on volumes ~64Tb and files approaching or larger than 1TB. However it does work! Just has the caveats that its rather slow!