I understand Teradata has a so-called fall back option. The idea is that data are stored twice; each record being stored to two different amps.
I saw a nice picture that descibes the situation. Each record (Row 1, Row 2 etc) are linked to an amp. It is shown in below scheme as yellow boxes. A seperate copy is made of all records. The copy is shown in an orange box. The records are all linked to another amp as the original record.
Having a fall back, certainly has advantages. If an amp fails, a query may use copies of the original records.
However, there is a flipside to this. It requires:
- more table space as copies must be stored
- more IO time as the copies must be written to disc
- more CPU time as he copies must be created by the system
Hence, this option is only used for critical tables where the additional costs are justified. Moreover, this option is advised to be used at table level only. This allows to make a fine distinction between tables for whih such additional costs are warrented and other tables where additional costs are not justified.