Repository Hardware for Instant Recovery Speed

[collinp]

If you were to build a new Veeam repository with HP DAS shelves connected to a DL360 Gen10, what drive type and controller type would you use to achieve the fastest possible Instant Recovery and restore speeds? Also, would you use Veeam dedup or Windows 2016 dedup? Would you choose NTFS or Refs? It needs to be reliable as well. We need about 300-500TB of DAS. Cost isn't an issue. Would you use the backup file per vm option?

Would you use SSD? I think SSD would help with the restore because the data is spread across the disk if you need to synthesize .vbk's and .vbm's. I don't think this would be sequential I/O where SAS is more beneficial. The data will need to tier to S3 or Azure.

In a nutshell, here are some examples of HP drive and controller types taken from the D3700 DAS Quickspecs (https://h20195.www2.hpe.com/v2/getpdf.a ... 227611.pdf):


HPE 3.84TB SATA 6G MU SFF SC DS SSD (Mixed Use)
HPE 3.84TB SAS RI SFF SC DS SSD (Write Intensive)
HPE 3.84TB SATA RI SFF SC DS SSD (Read Intensive)
HPE 900GB SAS 12G Enterprise 15K SFF
HPE 2.4TB SAS 12G 10K SFF
HPE 2TB SATA 6G Midline 7.2K SF

Would you do all drives of one type or mix and match using HP's smart cache feature on the P841?

Controller Types supported on D3700:
P421, P431, P441, P822 or the P841 Controllers

[HannesK]

Hello,
as you already mention, SSD is the fastest option. If money is no issue, go for 100% SSD (Mixed use)

The pure read / write speed seems to be a little bit slower (I did not do scientific tests and it was only around 3-5%) with ReFS, but for all merge / synthetic operations, it's of course much faster. For spinning disk storage with 500TB I would always go for ReFS. With all flash, well, NTFS could be "good enough".

I prefer not to use Windows deduplication because of filesize limits you would probably hit (just guessing by the size of repository you ask for).

Per-VM backup chains: yes, always!

Hope that helps,
Hannes

[collinp]

Has Refs stabilized? Will SSD wear out faster?

Also, what would help with the restore window taking a long time to come up?

Thank you!

Collin

[HannesK]

Hello,
about ReFS stabilization: for most customers yes. But you can still find some negative comments on ReFS. If that's not okay for you, then there is no need to think about. NTFS and All-Flash.

SSD wear leveling: out of scope from my side. Hardware vendor or Microsoft might know.

Sorry, I don't understand the last question

Best regards,
Hannes

[collinp]

What I meant by my last question is that when I open the FLR window, it takes forever to pull up the guest file system for restore.

I want the FLR window to pop up as quickly as possible. My guess is that in the background, Veeam has to synthesize the backup chain. We keep 90 days of daily backups. We perform active fulls monthly so we have 30 day backup chains. Would REFS or SSD disks help that synthesize that backup chain faster? I could perform weekly synthetic fulls to reduce the size of the backup chain to 7 days but then that sort of puts me down the *unreliable* path of using REFS because I would want my backup repository design to be efficient.

[HannesK]

Hello,
no, nothing is synthesized in the background for FLR. You won't see a relevant time difference. Only metadata is loaded and that's pretty fast (especially on SSD).

Best regards,
Hannes