Home > Storage > PowerFlex > White Papers > Dell APEX Block Storage for AWS: Backup and Recovery using DDVE and DD Boost Oracle RMAN Agent > Backup time and efficiency tests using BoostFS
In this test case, to simulate a generic application, we instantiated the Amazon EC2 c5n.9xlarge instance type that is used as a PowerFlex SDC instance with six PowerFlex volumes attached. The Amazon EC2 instance type i3en.9xlarge type with NVMe SSDs is used as a PowerFlex SDS instance. The Amazon EC2 compute optimized instance has total of 3 TB storage capacity and is populated with 1 TB of application data. The initial backup of the generic application uses 1 TB, and subsequent backups included a data change in a range of 10 percent (approximately 100 GB). The goal of the following tests is to show the different considerations when selecting the size of the EC2 instance for DDVE.
We obtained test results under the following categories:
The following figure shows the duration times of the incremental backups with the m5.xlarge instance type with 16 TB license model. The initial backup of all PowerFlex volumes took 136 minutes to complete reading all the volumes. The subsequent data backup of all PowerFlex volumes was completed in 92 minutes as there is only a 10 percent data change.
m5.xlarge instance is the default EC2 instance for the 16 TB license model.
Figure 18. Backup time using the BoostFS file system with the m5.xlarge instance type
The following figure shows the duration times of the incremental backups with the m5.8xlarge instance type with 16 TB license model. The initial backup of all PowerFlex volumes took 43 minutes to complete reading all the volumes. The subsequent data backup of all PowerFlex volumes was completed in 27 minutes as there is only a 10 percent data change.
These tests show the backup time benefits of using a more powerful EC2 instance for DDVE. Although m5.xlarge instance is the default EC2 instance for the 16 TB license model, the CloudFormation template provides an option to choose a higher instance type. As a result, the backup time was reduced.
Figure 19. Backup time using the BoostFS file system with the m5.8xlarge instance type
The following figure shows the initial backup of generic application data with the m5.xlarge instance type with 16TB license model. The initial backup is the most resource-intensive because the entire generic application data is transferred and protected on the DD system. The DD system CPU usage fell in all subsequent backups to a range between five to 10 percent. This change occurs because BoostFS processes only the unique data in each additional backup, freeing up CPU cycles for other operations.
Figure 20. CPU usage on the DD system with the m5.xlarge instance type
The following figure shows the initial backup of generic application data with the m5.8xlarge instance type with 16TB license model. The initial backup is the most resource-intensive because the entire generic application data is transferred and protected on the DD system. The DD system CPU usage fell in all subsequent backups to a range between five to 10 percent. This change occurs because BoostFS processes only the unique data in each additional backup, freeing up CPU cycles for other operations.
These tests show that the backup process is CPU intensive and DDVE can benefit from a more powerful EC2 instance. Although the incremental backups did not require high CPU utilization, the reduction in backup time shows the effect of the more powerful instance choice.
Figure 21. CPU usage on the DD system with the m5.8xlarge instance type
The following figure shows that the average initial full backup network bandwidth is 300 Mbps with the m5.xlarge instance type with 16 TB license model. Because all the data had to be sent to the DD system, the entire generic application data was considered as unique data to be protected on the DD system. The subsequent full backup network bandwidth rises significantly (copying the PowerFlex volumes parallelly), by a range between 220 Mbps to 250 Mbps, because the generic application data was already protected and only the unique data had to be transferred.
Figure 22. Network bandwidth on DD system using DD BoostFS with the m5.xlarge instance type
The following figure shows that the average initial full backup network bandwidth is 1700 Mbps with the m5.8xlarge instance type with 16 TB license model. Because all the data had to be sent to the DD system, the entire generic application data was considered as unique data to be protected on the DD system. The subsequent full backup network bandwidth rises significantly (copying the PowerFlex volumes parallelly), by a range between 2400 Mbps to 2500 Mbps, because the generic application data was already protected and only the unique data had to be transferred.
Figure 23. Network bandwidth on DD system using DD BoostFS with the m5.8xlarge instance type
The following figure shows that the initial size of data backup on the DD system is 3056 GB. After the data transfer to the DD system and generic application of deduplication and compression algorithms, physical storage consumption is reduced to 450 GB. The subsequent full backup of PowerFlex volumes’ data to the DD system, physical storage consumption is reduced to 65 GB. The amount was lower than the initial backup because deduplication became the dominant factor for the subsequent backups.
Figure 24. DD system pre-compression and post-compression values
The following figure shows that the initial backup consists of unique data and had the lowest total compression factor (6.8x). The subsequent backups were similar in the amount of unique data, and the compression factor increased accordingly. For the data reduction percentage, the initial backup had unique data and yielded the lowest reduction percentage. The subsequent backups were similar in the amount of unique data transferred and their reduction percentage increases accordingly.
Figure 25. Deduplication and compression savings