Best way to back up failover SQL Cluster without performance issues

[ibarizz]

Hello,

I have the following scenario and I would like to hear any recommendations or feedback you can provide.

I have a customer with multiple Failover SQL Clusters (2-3-4 and 5 nodes) in which their databases are configured in the SIMPLE recovery model. According to Veeam requirements, we take backups of these clusters using the Windows Agent Backup, and we have enabled application-aware backups to ensure consistent backups of their databases. Additionally, due to a customer requirement, we need to take backups every 8 hours, resulting in an RPO of 8 hours.

Every time we take a backup of our Failover SQL cluster with Veeam, they experience high CPU usage in their servers due to the Veeam Agent process. Sometimes the VSS freeze takes too long, which is unacceptable given the transactional nature of their databases. We have been unable to find a solution to this problem. So far, I have come to realize that it's normal to see this behavior.

That being said, my questions are:

1. Should we expect to see this kind of high CPU usage from the VeeamAgent processes?

2. If we decide to change the recovery model to FULL and only take backups once with periodic backups of transaction logs, could this potentially avoid the behavior we are currently experiencing?

3. So far, backup of transaction logs does not involve VSS, as databases do not need to be frozen, only information from the log archive needs to be gathered.
Will this task, when working continuously in a Failover Cluster where we use agents to communicate with Veeam, increase CPU usage for the SQL Servers in the same manner as it does during full or incremental backups?


Thank you in advance, and I apologize for the lengthy text. I just wanted to check if somebody can confirm this or perhaps offer a different solution to this issue. In my opinion, transitioning to the FULL recovery model and taking backups of transaction logs will reduce the CPU load on these SQL Servers.

Cheers.

[Mildur]

Hello Ibarizz

1. Should we expect to see this kind of high CPU usage from the VeeamAgent processes?

The Veeam Agent process does the entire processing such as reading data from disk and compressing the data. This can lead to a high CPU load. If you feel the CPU usage is too high, please contact our customer support team and let them check the logs and performance data. They can also provide a recommendation to optimize the performance of your backups after investigating the log files.

2. If we decide to change the recovery model to FULL and only take backups once with periodic backups of transaction logs, could this potentially avoid the behavior we are currently experiencing?

I expect lower CPU usage compared to creating image level backups. Also transaction log backup may be faster because the agent doesn't have to wait for VSS snapshot to be created. For databases with a huge load it may take some time for the VSS snapshot to be created.

3. So far, backup of transaction logs does not involve VSS, as databases do not need to be frozen, only information from the log archive needs to be gathered.
Will this task, when working continuously in a Failover Cluster where we use agents to communicate with Veeam, increase CPU usage for the SQL Servers in the same manner as it does during full or incremental backups?

No, it shouldn't increase CPU usage.

Another option you could test is our Veeam Plug-in for Microsoft SQL Server:
https://helpcenter.veeam.com/docs/backu ... ml?ver=120

Best,
Fabian

[PetrM]

Hello,

Adding my 2 cents: one of the primary goals of the SQL plug-in was to provide high performance and scalability. In our environments, we successfully tested up to 2000 DBs per server/cluster. I recommend trying it and please share the feedback, all the requests and ideas are always welcome.

Thanks!

[ibarizz]

Thank you PetrM for your answer.
We are going to try taking backups with recovery mode in full as during night times our customer wouldn't care so much about the cpu usage of their databases.

Using SQL plug-in ain't the best option at all as we would take back up of the database, but not the SQL engine. Anyways, we will take that option in consideration as well.

Thank you again for your quick answer.