EMC VNX

How will clients be affected during Data Mover failover

NFS Clients: NFS is a stateless protocol.A stateless server does not keeps state between connections.NFS version 3 is the standard for NFS and is the most stable and mature version of the NFS protocol. NFS clients will likely see a “NFS Server not responding” message when failover is underway. NFS operation will continue again when failover is completed.

CIFS Clients: A stateful server keeps state between connections. If no file writes are taking place during the time that the Data Mover is failing over CIFS users may not see anything. If a CIFS user tries to start a write while the Data Mover is failing over, the user will see an error and will be prompted to save the file elsewhere. If the user waits a few seconds and re-attempts the save, the user will be successful. If the user was in the middle of a save when the failover starts, the user will get an error, but should be able to restart the save after the failover is complete.

All clients will be affected by Data Mover failover. The question is not will they be affected, but how they respond and can the application tolerate the interruption.

NFS:

• An NFS client that does not issue any I/O to the Data Mover during the failover will see no impact at all and no change to cope with.

• An NFS client that does issue I/O to the Data Mover during the failover will not see any response to that I/O. NFS clients will simply wait and periodically retry I/O in these circumstances. When the failover is complete the I/O will succeed and the NFS client will carry on. It will not see any changes that require any other recovery procedures. Note that although the NFS client will recover fine, applications that are relying on that NFS data store will react in accord with their individual design specifications or administration parameters.

CIFS:

• A CIFS client that does not issue any I/O to the Data Mover during the failover will have its connection to the Data Mover dropped. CIFS clients are used to this happening if they have not accessed a share for a time, so they automatically re-establish their share connection when required.

• A CIFS client that does issue I/O to the Data Mover during a failover will timeout and return an error to the application after either 25 or 45 seconds depending on the state of the I/O. Again, how the application handles this error is dependent on the application. In addition to the I/O when the client does reconnect after the failover is complete it must re-establish its connection.

• It is important that the System Administrator to be aware of applications sensitive to extended timeout errors and to schedule maintenance during low activity periods. A scheduled failover performs an orderly shutdown that is time intensive and refuses new connections while failing over. This overhead is not experienced to an equal degree during a genuine Data Mover panic and automatic failover.