Replication policies and jobs

Data replication is coordinated according to replication policies and replication jobs. Replication policies specify what data is replicated, where the data is replicated to, and how often the data is replicated. Replication jobs are the operations that replicate data from one Isilon cluster to another. SyncIQ generates replication jobs according to replication policies.

A replication policy specifies two clusters: the source and the target. The cluster on which the replication policy exists is the source cluster. The cluster that data is being replicated to is the target cluster. When a replication policy starts, SyncIQ generates a replication job for the policy. When a replication job runs, files from a directory tree on the source cluster are replicated to a directory tree on the target cluster; these directory trees are known as source and target directories.

After the first replication job created by a replication policy finishes, the target directory and all files contained in the target directory are set to a read-only state, and can be modified only by other replication jobs belonging to the same replication policy. We recommend that you do not create more than 1,000 policies on a cluster.

Note Image

To prevent permissions errors, make sure that ACL policy settings are the same across source and target clusters.

You can create two types of replication policies: synchronization policies and copy policies. A synchronization policy maintains an exact replica of the source directory on the target cluster. If a file or sub-directory is deleted from the source directory, the file or directory is deleted from the target cluster when the policy is run again.

You can use synchronization policies to fail over and fail back data between source and target clusters. When a source cluster becomes unavailable, you can fail over data on a target cluster and make the data available to clients. When the source cluster becomes available again, you can fail back the data to the source cluster.

A copy policy maintains recent versions of the files that are stored on the source cluster. However, files that are deleted on the source cluster are not deleted from the target cluster. Failback is not supported for copy policies. Copy policies are most commonly used for archival purposes.

Copy policies enable you to remove files from the source cluster without losing those files on the target cluster. Deleting files on the source cluster improves performance on the source cluster while maintaining the deleted files on the target cluster. This can be useful if, for example, your source cluster is being used for production purposes and your target cluster is being used only for archiving.

After creating a job for a replication policy, SyncIQ must wait until the job completes before it can create another job for the policy. Any number of replication jobs can exist on a cluster at a given time; however, no more than 50 replication jobs can run on a source cluster at the same time. If more than 50 replication jobs exist on a cluster, the first 50 jobs run while the others are queued to run.

There is no limit to the number of replication jobs that a target cluster can support concurrently. However, because more replication jobs require more cluster resources, replication will slow down as more concurrent jobs are added.

When a replication job runs, OneFS generates workers on the source and target cluster. Workers on the source cluster read and send data while workers on the target cluster receive and write data. OneFS generates no more than 8 workers per node per replication job. For example, in a five node cluster, OneFS would create no more than 40 workers for a replication job.

You can replicate any number of files and directories with a single replication job. You can prevent a large replication job from overwhelming the system by limiting the amount of cluster resources and network bandwidth that data synchronization is allowed to consume. Because each node in a cluster is able to send and receive data, the speed at which data is replicated increases for larger clusters.