Replication Options for Isilon Shares


Replicating filesystems on the Isilon

CNS has three Isilon arrays:  one located in the SSRB datacenter, one in the UFDC datacenter, and one in Atlanta.

At this time, the Isilon in Atlanta is intended solely as a Disaster Recovery (DR) resource - it exists as a replication target only, and will not be available to serve active content in case of a disaster because of the amount of bandwidth available from UF to Atlanta. In case of a catastrophe, content that is mirrored in Atlanta will be migrated back to UF as soon as local infrastructure has been made available.

Connectivity from the UFDC array is via redundant fibre connections that take separate paths from the UFDC to campus. Expectations are that the Isilon should roughly mirror performance available from the array hosted in the campus datacenter.

At this time, Bronze level file hosting services from CNS do not include a replication option. Silver level service includes replication between one of the local arrays (based in either SSRB or UFDC) to the Atlanta off-site array, and Gold level service includes replicating between both of the local datacenters, and an off-site DR copy in Atlanta.

Replication in the Isilon infrastructure acts in an active/passive fashion. The active copy of the filesystem is read/write, and the replicated copy of the filesystem is read-only. Takeover of the filesystem is not automatic - we will need to tell the read-only copy of the filesystem to become the active copy, and the DNS record for the file server name will need to be changed.

Replicated filesystems exist in different ip address pools (one for each cluster), so it's important to configure ACLs for each pool identically when you make an ACL addition or change.

We recommend that Windows-centric departments make use of DFS and add both filesytems as targets for the share address. In case of a failure, there will be a small window where the available filesystem will be read-only until the switch to change the replication target to an active copy is thrown, but no other configuration changes should be necessary other than that. If you do not already have a DFS infrastructure, you can use UFAD's DFS infrastructure if you wish. Please contact support@ad.ufl.edu for more information about this, or open a service request at https://my.it.ufl.edu to start the process.

In addition, for Mac and Linux hosts that connect to the filesytem directly, during the provisioning process we will add a CNAME address that points to the currently active filesystem. This will need to be updated manually in case of a failover, but we hope to have a faster way to automate this process via the Bluecat to try to minimize the amount of time that it takes to update the record.