Introduction to Data Management

Ensuring Database Consistency With TMF
15873 Tandem Computers Incorporated 7-7
Protection Across the Network
Together with the EXPAND software, TMF supports consistency across all portions of
a database distributed over a network. This consistency prevails no matter where a
transaction originates or how it is geographically distributed. As an example, a
transaction might produce updates to files stored at three different network nodes. If
the updating required at one node cannot be completed, TMF backs out every part of
the transaction that was begun at other nodes before the transaction failed. This
action restores the database to its original state on all nodes participating in the
update.
TMF maintains separate audit trails on each node. Data images of changes to a
database file located at a particular node are written to an audit trail file on that same
node.
During the update operation, but before the records are unlocked, TMF on the node
where the transaction originated communicates with all other nodes involved in the
transaction to ensure the other nodes can permanently commit the change to the
database.
If any participating node cannot commit the transaction, either because the node
cannot be reached or for some other reason, record locks are not released and TMF
backs out the transaction on all nodes where the transaction was implemented.
Backout occurs independently at each node.
After a distributed transaction is committed on all nodes involved, the record locks
are released. Locks are released independently on each node; if a participating node
cannot be reached during the release operation, locks are released on all reachable
nodes—and when communication is restored to the unreachable nodes, locks are
released on those nodes, too.
High OLTP Performance
Even in demanding OLTP environments involving thousands of concurrent
transactions, TMF sustains support for high-volume OLTP. During updating
operations, TMF writes the before-images and after-images to the audit trail before it
commits the transaction to the database. This write-ahead capability reduces I/O
activities, promoting high transaction rates and fast response times. TMF also allows
you to commit a group of transactions with a single disk write operation, further
reducing the number of I/O operations required.
Protection Against System
Failures
Although the failure of an individual transaction can be resolved by transaction
backout, the more severe problem of data loss due to system failure requires a
different solution. In a system failure, the entire computer system shuts down,
potentially resulting in a serious loss of data. Sometimes this loss can be very
widespread.
System failures can arise from a variety of causes, including multiple hardware
failures, extended local power failures, or even city-wide power outages. When this
kind of failure occurs, TMF allows you to restore your database to a consistent state.