Logging speed is a bottleneck for IMS performance, and the WADS I/O has been a bottleneck on logging speed. The enhancements in IMS Release 15 to exploit System z High Performance FICON (zHPFâ„¢) and zHyperWriteâ„¢ reduce the latency for writing to the WADS data set, increasing logging speed.
IMSâ„¢ writes log records to a DASD data set called the online log data set (OLDS).
The OLDS is made up of multiple data sets written in wraparound form. Using more than one OLDS enables IMS to continue logging when the first OLDS is full. Also, if an I/O error occurs while writing to an OLDS, IMS can continue logging by isolating the OLDS where the problem occurred and switching to another one.
IMS can write committed log records to the write-ahead data set (WADS) so that these records are externalized to avoid the need to write partially filled and padded log blocks to the OLDS.
When the OLDS is full, it is archived to the system log data set (SLDS). How frequently the OLDS is archived is controlled by the administrator. The system reuses the OLDS after it has been archived. An SLDS can be on DASD or on tape. The contents are used as input to the database recovery process.
To enhance performance and to optimize space on OLDS, incomplete or partially filled buffers are written to a WADS when necessary for recovery. The WADS are high speed DASD data sets with a high write rate. Only complete log buffers are written to OLDS. When the log data is on an OLDS, the equivalent WADS records are ignored.
The log write-ahead function ensures that all log records are on the log before IMS writes changes to a database in any of these situations:
- When IMS needs to reuse the databases buffer (if this is before commit)
- During commit
- During VSAM background write
IMS Release 15 Enhancements for WADS Performance
The IMS Release 15 WADS I/O enhancements allow clients to exploit the zHPF and zHyperWrite capability provided by IBM DS8000â„¢ Storage Systems. The zHPF feature provides improved channel and control unit efficiency, lower latency and improved reliability/availability as compared to FICON. The zHyperWrite feature is able to mitigate the synchronous replication delays that occur when using Metro Mirrorâ„¢ technology for continuous availability and disaster recovery.
The experiment below shows a typical IMS work load run during the IBM testing of IMS Release 15. The IMS WADS I/O service time is shown with zHPF and zHyperWrite turned off and then again with zHPF and zHyperWrite turned on. Up to a 50% reduction in WADS I/O service time was observed.
Writing to the IMS WADS dataset is a critical path for IMS transaction performance. Exploitation of the enhancements in IMS Release 15 can reduce the I/O service time for WADS writes by as much as 50%.