RDF System Management Manual

Managing RDF
HP NonStop RDF System Management Manual—524388-003
5-35
Exception File Optimization
Exception File Optimization
The RDF exception files reside in the control subvolume on $SYSTEM. The name of
each is the name of the updater’s primary system volume.
Each updater maintains an exception file in which it identifies every audit record that
must be undone on the backup database during a takeover. Typically records must be
undone because the outcome of the associated transaction is unknown. When
protecting auxiliary audit trails, however, the outcome of a transaction might be known
(a COMMIT record is present in the Master Image Trail), but there is no way of
knowing for certain that all the associated audit records were successfully replicated to
the auxiliary image trail prior to the takeover.
If you are protecting only MAT volumes, the amount of undo required during a takeover
is usually small. If one or more long-running transactions are active at the time of a
takeover, however, the amount of undo required can increase substantially (depending
upon the amount of audit records generated by those transactions).
If you are protecting auxiliary audit trail volumes, a considerable amount of undo could
also be required if any of the extractor-receiver pairs (master or auxiliary) falls behind
the others prior to a takeover.
If you have configured an RDF network to replicate network transactions, a
considerable amount of undo could also be required if any of the nodes in the network
falls behind the others prior to a takeover.
In any case, if an updater has a large number of audit records to undo during a
takeover, the performance of its undo pass is negatively affected by logging exception
records. Therefore, the manner in which exception files are used is a configurable
attribute.
To set this attribute, use the following RDFCOM command:
SET RDF UPDATEREXCEPTION {ON | OFF}
When this attribute is set to ON (the default value), the updater logs an exception
record for every audit record it must undo during a takeover.
When this attribute is set to OFF, the updater logs exception records only for the first
and last audit records that must be undone (the minimum logging necessary to support
Triple Contingency operation).