one LUN or multiple LUNs

[apolloxm]

Hi,

Can someone tell me that Is there any difference performance between one LUN and Mutiple LUNs. for example, We had 6 TB free space. Is better to have 2 LUNS or one 6TB LUN as repository. Those repositories will be formated as REFS with 64K to support Veeam fast clone. but I don't know that is there any performance difference. We will have backup jobs. after the backup jobs finished, we will start to offload the data to cloud. Thanks!

[Mildur]

Hi

6TB seems not that much storage. I would just use one LUN to keep it simple. If you have multiple LUNs, you have to use a SOBR with multiple extends. And one LUN should have 2-3x times the size of the biggest full backup file.
But without knowing your environment, it's difficult to give a recommendation.

Thanks
Fabian

[apolloxm]

Hi

6TB seems not that much storage. I would just use one LUN to keep it simple. If you have multiple LUNs, you have to use a SOBR with multiple extends. And one LUN should have 2-3x times the size of the biggest full backup file.
But without knowing your environment, it's difficult to give a recommendation.

Thanks
Fabian

Hi,

Thanks for your reply! 6 TB just an example. Currently, we totally have 33 TB for one Week Full backup and set the retention policy to 28 days. so at least will have 4 weeks or 5 weeks data in totally. 33*5=165TB data in totally, Can you give us a recommendation how many luns should I created? how did you caculate that?also,we tototally have 41 jobs. some of the jobs one VBK file will reach to 2 or 3 TB.

[Mildur]

If you use reFS, you won't have 33*5=165 TB Data on the repository. reFS with FastClone will store 1 Fullbackup and then only changed blocks (if you do synthentic fulls). I assume it will be around 40-42 TB.
Multiple LUNs at different repository servers can help you to distribute the load of the backup repository tasks.

For a recommendation, I still missing information:
- How many repository Server will be used?
- What storage appliance do you have?
- How many vms will be backed up?
- How is the LUN connected to the repository server? iSCSI, FC, SAS?

I would probably use just 1 LUN even with a 50TB repository.

[apolloxm]

- How many repository Server will be used?
We installed Veeam Server all in one.
- What storage appliance do you have?
We had two QNAP Storage
- How many vms will be backed up?
We had 72 VMs with hyper-V will be backuped up.We also are using Veeam agent to Backup Virtual machine with ISCSI connection.
- How is the LUN connected to the repository server? iSCSI, FC, SAS?
ISCSI Connection.

Multiple LUNs at different repository servers can help you to distribute the load of the backup repository tasks.

Except for this, How about the disk I/O Performance and restoring speed ? Is there any difference.for example, Can you tell me the difference performance between one LUN with 40TB size with 8 concurrent task and two LUN with 20TB size with 4 concurrent task per lun?

Currently, We are using SOBR with performance policy. We had 10 LUNs, 9 LUNS with 20 TB per lun and 1lun have 70TB.Base on my experience, Veeam always try to select extent from the group with highest priority with greatest amount of free space.so veeam always choose the 70TB as the backup reposistory. it looks like the performance on the 70TB not good than other LUNs. As we are using the performance policy and run active full for weekly backup.so, fast clone doesn't work for us. We also had backup copy jobs was running with active full by weekly, so the performance of Veeam server is not good. We are thinking to redesign our Veeam Backup Repository. but we don't know how many luns should be created. Thanks!

[Mildur]

Can you tell me the difference performance between one LUN with 40TB size with 8 concurrent task and two LUN with 20TB size with 4 concurrent task per lun?

I would say, it depends on the amount of disks and their I/O capability if both LUN's are stored on a single QNAP. Max I/O value from one hard disk doesn't get higher if you use two LUN's instead of one.
The performance optimization I expect is when you connect each LUN to different backup repository machines. Each machine can act separately with the LUNs and do not share CPU and Memory for all extends. Also you could benefit from the higher network throughput, if your QNAP has multiple NICs and your repository server are running on different hosts.

- How many vms will be backed up?
We had 72 VMs with hyper-V will be backuped up.We also are using Veeam agent to Backup Virtual machine with ISCSI connection.

At my workplace before I joined veeam, we had one project with a single extend for 150 machines. Worked great.
But it was a physical server with locally attached Disks.

We are using SOBR with performance policy. We had 10 LUNs, 9 LUNS with 20 TB per lun and 1lun have 70TB.Base on my experience, Veeam always try to select extent from the group with highest priority with greatest amount of free space.so veeam always choose the 70TB as the backup reposistory. it looks like the performance on the 70TB not good than other LUNs.

Yes, with weekly active fulls, backups from a single machine can switch the extend each week. Any reason why you are not using synthetic Full with Fast Clone? Is it because QNAP could be unsupported for reFS from a Microsoft view?
Synthetic Fulls with FastClone would improve this situation.

[apolloxm]

I would say, it depends on the amount of disks and their I/O capability if both LUN's are stored on a single QNAP. Max I/O value from one hard disk doesn't get higher if you use two LUN's instead of one.
The performance optimization I expect is when you connect each LUN to different backup repository machines. Each machine can act separately with the LUNs and do not share CPU and Memory for all extends. Also you could benefit from the higher network throughput, if your QNAP has multiple NICs and your repository server are running on different hosts.

Answer: Per my testing with DiskSPD, 40TB size and 20TB size disk I/O speed almost the same. Is that means 2 LUNS with 20 TB size disk speed I/O are faster than 40 TB LUN size?


Yes, with weekly active fulls, backups from a single machine can switch the extend each week. Any reason why you are not using synthetic Full with Fast Clone? Is it because QNAP could be unsupported for reFS from a Microsoft view?
Synthetic Fulls with FastClone would improve this situation.

Answer: We used synthetic full in the past with SOBR performance policy, since the files was hosted on the difference extend and We also tested with active full. it looks like the performance with active full is better than sythetic full. another reason, We did some maintaince to our SharePoint database server(cut database from one into to ) in the past.and we didn't stop the backup job with synthetic full, at the end, we can't restore the Sharepoint data from Veeam, We contacted this with your support team, but they can't find the root cause and we fixed the issue by creating an active full, that's why we thought active full is better than synthetic full.

[apolloxm]

if i choose one lun as repository how many concurrent task should i limit

[apolloxm]

anyone can help me？

[Mildur]

Hi Apollo

It depends on some factors.
If you already have the system in place, you can start with the recommended max concurrent task limit for your CPU/Memory of the repository machine. Monitor your repository and LUN performance and adjust the limit.

Description and resource requirement of repository tasks:
https://helpcenter.veeam.com/docs/backu ... positories

You can find a generic overview of the required I/O's for backup processes here:
https://bp.veeam.com/vbr/4_Operations/O ... -processes

How many VMs (and VM disks) will you write to this LUN? I assume you will use Per-VM backup chains?

Best,
Fabian