CloudPools is designed to move cold data from primary storage to the cloud. It is deliberately slow to ensure that it does not compete with things that are performance sensitive such as SMB and NFS user activity. By default, CloudPools is using 10 threads per node, which balances CloudPools CPU usage with other cluster functions. Using the default number of threads is recommended for typical workloads. CloudPools does provide an option to modify the number of archive and recall threads. Modifying the number of archive and recall threads can improve archive and recall performance but can also have a significant impact on the CPU load of your system.
Note: Contact your Dell representative if you want to configure a higher number of threads.
CloudPools archive performance and recall performance are highly dependent on many factors, such as the network bandwidth between the PowerScale cluster and the cloud, available system resources, and file size. These performance considerations include:
- As the file size increases, the CloudPools archive and recall performance also increases. However, the effect on archive performance is minor when the file size is greater than or equal to 10 MB. The effect on recall performance is negligible when the file size is greater than or equal to 10 MB.
- As the thread counts increase, the CloudPools archive and recall performance also increases. However, the effect on archive and recall performance is negligible when the number of threads per PowerScale node is greater than or equal to 40.
- For a single large file, the effect on archive and recall performance is negligible regardless of the file size or number of threads per node. A single thread manages a single file transfer on a single node.
- Starting from OneFS 9.3.0.0, CloudPools creates Likewise sparks to drive the read of each CDO from the cloud. This enhancement can cache multiple chunks or CDOs of a stub file concurrently to improve CloudPools read and recall performance.
- The effect on archive and recall performance is negligible regardless of the number of file pool policies or jobs.
- With the expansion of PowerScale nodes, CloudPools archive and recall performance increases, but not linearly.
- For a single, heterogeneous cluster, tier 1 (fast) node pool has a minor impact on CloudPools archive performance and a large impact on CloudPools recall performance. The archive and recall performance are better when data is stored in the tier 1 node pool. The setting Data Storage Target of a file pool policy can determine the node pool for recall. However, the node pool cannot be changed for inline read. The node pool for a stub is used for inline read for this stub.
- Not All Nodes on Network (NANON) cluster has a large impact on CloudPools archive and recall performance.