Data Protection

Fallback

Fallback protects your data by storing a second copy of each row for a table on an alternative "Fallback VPROC."  If a VPROC fails, the Teradata system accesses the Fallback VPROC/rows to meet the request.  This way Teradata provides fault tolerance at the data level.  With fallback tables, if one VPROC fails, all the data is still available to the users.  This allows the users to continue using the database and the VPROC to be fixed without a system outage.  

Fallback is not always necessary on the 5100 systems by the use of disk arrays.  If the VPROCs is combined with disk arrays, the need for fallback is lessened.  There are two reasons for this:  1)  VPROCs do not fail because the PDE restarts the VPROC with a Teradata Reset, 2)  RAID technology protects the data on the disks.  If a disk goes down, the data is either available from the redundant copy (RAID1) or can be reconstructed (RAID5).

Cost and Benefits of Fallback

There are certain costs and benefits associated with Fallback on the Teradata Database System.  The costs of using Fallback include requiring twice the disk space for storage, and twice the I/O for inserts, updates, and deletes (all fallback I/Os are performed in parallel to the primary I/O).  The benefits of Fallback are protecting your data from hardware failure (VPROC or DSU), protecting your data from software failure, and automatic recovery after repair is complete, minimum recovery time after repairs are complete, and requires no extra I/O for the SELECT SQL function.

Clustering

A Cluster is a group of VPROCs that provide fallback for each other.  Clustering does not have any effect on the data distribution of the primary rows of the table.  Regardless of which VPROC the primary row hashes to, the fallback row will always go to a different VPROC in the same cluster.  Cluster sizes are set through a Teradata console utility and may range from 2 to 16 VPROCs per cluster.   When one VPROC in a cluster fails, the data will be accessed on the fallback VPROC on the same cluster.  While the one VPROC is down, the remaining VPROCs must remain up and on-line.