Copy Performance from a Recovery Point is Slow (4033540)

Copy speeds from a recovery point can vary significantly due to multiple environmental factors. Understanding these variables can help improve performance and ensure efficient data recovery processes.

When restoring data from a mounted recovery point, several factors can impact copy performance:

1. Speed of the Storage Array:

   • The speed of the storage system where the repository resides directly influences restore times. Faster storage devices allow quicker data retrieval and copying.

2. Repository Space and Fragmentation:

   • Repositories can become full due to unexpected base images, low deduplication cache size, high change rates on protected machines, or simply reaching their useful limits. A nearly full repository will impact job performance as the Data Virtualization Module (DVM) must process more data to locate records for overwriting or deletion. It is recommended to maintain at least 20% free space in the repository.

   • The longer a repository exists, the greater the likelihood of fragmentation. A full repository significantly increases this probability. Moreover, the fragmentation in a full repository does not disappear when space is freed. This persistent fragmentation adversely affects performance.

   • Recommended steps to defragment backup data include:

     1. Archiving or replicating the current protected data.

     2. Deleting the current repository.

     3. Creating a new repository.

     4. Importing the archived/replicated data.

   • Another option is to run the repository optimization job. This requires increasing the deduplication cache size, which may necessitate checking if the Core server has sufficient RAM to support the larger cache. For more details, refer to the Understanding the Repository Optimization Job article.

3. Concurrent Repository Operations:

   • Pause other jobs running on the repository to reduce I/O contention and improve performance for the current restore operation.

4. Network Speed (if applicable):

   • When copying data to a network location, the speed and reliability of the network connection can significantly impact restore performance. Optimizing network infrastructure or choosing a faster transfer medium can improve copy speeds.

   • Assess the network transfer rate between the Core and the target machine using tools such as iPerf to identify potential bottlenecks. Refer to the How to Run iPerf to Collect Bandwidth Statistics Between Machines article for detailed instructions.

5. Understanding Transfer Rates in Rapid Recovery:

   • When performing data recovery, customers may notice that the transfer rate appears slower than expected. This is because Rapid Recovery employs random read operations when accessing data from the repository. Random reads inherently involve frequent disk seek operations, which can impact performance on traditional storage systems.

   • In contrast, sequential read operations, which access data in a continuous block, typically result in faster data transfer rates because they minimize the mechanical movement of disk heads. While sequential reads may exhibit a higher transfer rate, the design of Rapid Recovery prioritizes data integrity and versatility, requiring the use of random reads to meet these goals.

   • Additionally, Rapid Recovery is a block-level backup solution. When copying files from a mounted recovery point, the data presented in the mount is not preloaded as complete files but is instead retrieved on-demand as blocks from the repository. This process uses RAM to manage the mounted recovery point while accessing the necessary data from the repository storage.

   • We recommend ensuring that the storage subsystem is optimized for random I/O workloads to maximize performance under these conditions. For enhanced performance, consider using SSDs or storage arrays optimized for random read/write operations.

Recommendations:

• Assess and, if necessary, upgrade the storage infrastructure hosting the repository to reduce bottlenecks. Ensure the repository storage supports the amount of random IOPS recommended by the Rapid Recovery Sizing Guide.

• Store repository metadata on a faster storage tier, such as SSDs, to improve performance.

• Verify that the current storage and network drivers are up to date to avoid potential performance issues.

• Pause other jobs running on the repository to reduce I/O contention and improve performance for the current restore operation.

• Evaluate and upgrade network components, such as switches and cables, to improve transfer rates when restoring to a network destination.

• Assess the network transfer rate between the Core and the target machine using tools such as iPerf. Refer to the linked article for more details.

• Review and adjust Rapid Recovery Core-level performance settings to optimize operations as outlined in the Understanding Rapid Recovery Core-Level Performance Tuning Settings article.

Referenced Links:

• Rapid Recovery Sizing Guide

• Understanding Rapid Recovery Core-Level Performance Tuning Settings

• Understanding the Repository Optimization Job

• How to Run iPerf to Collect Bandwidth Statistics Between Machines