TMF Introduction (G06.24+)

Application Performance
HP NonStop Transaction Management Facility (TMF) Introduction—522414-001
4-5
Writes to Audit Trail Files
Writes to Audit Trail Files
TMF protects transactions and the integrity of the audit trail while optimizing the
performance of writing to audit trail files.
TMF provides audit-writing procedures that minimize physical access time. It does this
by writing audit records for multiple transactions at the same time. This is called group
commit. In addition to commit records, all accumulated audit records are written to
disk by the audit trail disk process, regardless of the transaction to which they apply.
Group commits minimize the impact on performance caused by the additional
input/output required for auditing files under TMF protection.
Sequential writes are much more efficient than random writes. For this reason, TMF
makes sequential writes to the audit trail rather than the random writes typically used
for accessing data volumes.
In an application protected by TMF, even though transactions might access many
volumes on a node, the writes go only to the active audit volume. The single write to
the audit trail ensures that the database changes to files are permanent, even if the
database is distributed on many volumes of a node.
Performance Versus Transaction Rate
TMF increases application performance as the transaction rate increases, because of
the effects of buffered input/output for data volumes and group commits to the audit
trail.
Disk writes often occur at specific intervals, based on system events that cause
physical writes from cache to disk. As the number of transactions per interval
increases, throughput increases.
For example, compare throughput of records to disk with throughput of people on a
subway. Trains leave the station every 15 minutes. During a time of heavy use, many
more people are transported than during a time when only a few people want a ride.
The trains are full during heavy use and nearly empty during light use.
A comparison of throughput depending on transaction rate is shown in Figure 4-4. For
the comparison, assume a running application where all the transactions and their
input/output requirements are similar. The input/output for the comparison represents
writes for TMF transactions on audited files with buffered writes. At certain intervals,
cache for a database disk is written and includes updates for that disk from several
transactions. With the slow transaction rate, updates for only a few transactions are
written. With the higher rate, updates for thousands of transactions are written.