Hardware and Performance for Hardened Repository

[ElmerAcme]

Hello,
I want to implement a Hardened Linux Repository and got some questions for the right sizing…
Actual setup:
VMWare Cluster, Fullflash SAN, Hardware Veeam Server, 2x Dedupe Appliances as backup repository, FC TapeLibrary with 2x LTO8
Fullbackup Size about 100 TB, Backup from SAN, up to 2x 1,1 GByte/s backup performance.

Backup to the two dedupe appliances is pretty fast. On the other side, the tapeout and restore performance is pretty slow, with about 150 MByte/s max.
As far as I know, this is the normal behavior of those appliances.

The Hardened Linux Repository should solve the problem for immutable backups and a faster restore.

Now I’m struggling to find right hardware for this.
The Veeam sizing tool only points to CPU, RAM and Size (240TB !), but not to performance.

At this size there will be a lot of big 7.2k 3,5” Drives. Can a backup be as fast as to the dedupe appliances?
What speed can I expect for backup and restore?
Hardware RAID or software RAID?
Did someone have a personal hardware recommendation for me? Normally we use DELL hardware.

Anything else to know?

Thanks a lot

[HannesK]

Hello,
I'm trying to understand your "!" for the disk size... which Veeam sizing tool did you use, which values did you put in and what does the "!" mean?

Yes, dumb 7.2k disks are much faster with the same number of spindles than a dedupe appliance because the whole dedupe overhead does not exist. Over time, speed will go down a bit because of fragmentation with block cloning. But it will always be better than with a dedupe appliance.

I would always go for hardware RAID. If you want to go faster than 2GByte/s, I recommend to talk to your server vendor because there seem to be some RAID controllers that have limitation there (two RAID controllers would be the workaround then)

You can use any server vendor you prefer. I'm sure that DELL has something (server + shelves or internal disks depending on how fast you want to go). HPE Apollo are probably the most popular servers for Veeam backup repositories if high density is needed.

Cisco and HPE posted some performance values and you can translate that to any vendor because the difference in RAID controllers, mainboards etc. is not that huge
https://community.hpe.com/t5/around-the ... -p/7146835

Apollo 4510 Gen10: 56x18TB Drives – 864TB @ 4 rack units
Backup 12 GiB/s = 42 TiB/h
Restore 9 GiB/s = 31.6 TiB/h

Apollo 4200 Gen10: 24x18TB Drives – 360TB @ 2 rack units
Backup 6 GiB/s = 21 TiB/h
Restore 4.5 GiB/s = 15.8 TiB/h

DL 345 Gen10: 12x18TB Drives – 162 TB @ 2 rack units
Backup 3 GiB/s = 10.5 TiB/h
Restore 2.25 GiB/s = 7.9 TiB/h


https://www.cisco.com/c/en/us/td/docs/u ... veeam.html

UCS S3260 M5: 56x NL-SAS Drives @ 4 rack units
Backup 5.5 GB/s = 19.8 TB/h
Restore 3.2 GB/s = 11.2 TB/h

UCS C240 M5: 24x SSD Drives @ 2 rack units
Backup 3.2 GB/s = 11.5 TB/h
Restore 4.2 GB/s = 15.1 TB/h



Best regards,
Hannes

[ElmerAcme]

Hello,

the ! should remind me for adding some more information…
(! = For a 14 Day retention of 100TB, I should use 240TB. So, if the filesystem does block cloning, it seems a bit too much.)

Thanks for adding the throughput performance.
About 2 GByte/s for read and write would be enough.
Didn't expect that those drives have such a troughput.

I will get in contact with DELL.

Many thanks

[HannesK]

Hello,
ah, okay

the performance values above are for full backup / full restore on an empty system. That's why I mentioned, that they will go down over time due to fragmentation / block cloning. I would calculate 60-80MByte/s per disk for restore to be a bit more conservative. I would also not go for the largest disks available, because of RAID rebuild times.

For a 14 Day retention of 100TB, I should use 240TB. So, if the filesystem does block cloning, it seems a bit too much

Agree. And https://calculator.veeam.com/vbr/ gives much lower values depending on how much growth and change rate I put in.

Best regards,
Hannes