As we continue to wind our way through the multiple improvements that DB2 V9 brings to the IBM utilities, it is time to address the enhancements made to the REORG utility.
First up, let’s talk about performance improvements. In DB2 9 for z/OS, the REORG utility can unload and reload partitions in parallel. This should result in a nice reduction in elapsed time when you are reorganizing partitioned table spaces. To enable this improvement you will need to code the NOSYSREC keyword or the UNLDDN keyword with a template. Note that NOSYSREC is always used for SHRLEVEL change.
Parallelism will not be enabled if any of the following conditions apply:
- DATAWKnn ddnames are specified in the JCL
- SORTDEVT keyword is not specified
- UTPRINT is allocated to anything other than SYSOUT
- The REBALANCE keyword is used.
Also consider that REORG SHRLEVEL CHANGE can use subtasks during the LOG phase to speed up the processing of log records. Subtasks will be used when unload and reload have been done by partition and there are log records to apply for the partition.
BUILD2 Phase Eliminated in Online REORG
As you probably know, an Online REORG uses shadow data sets. The reorganization is done to the shadows while the data remains available in the primary data sets. When the REORG is done, the shadow data sets are switched to be the primary data sets.
As most DBA know, reorganization by partition can create a long outage when non-partitioned indexes (NPIs) exist on the table. Why is this so? Well, prior to V9, during Online REORG, NPIs had to be updated with the new RIDs for the data of the partition being reorganized during the BUILD2 phase. And that is what typically caused the sometimes significant outage.
With DB2 V9, the BUILD2 phase is eliminated. Instead, the entire NPI will be reorganized and then switched in the SWITCH phase along with all of the other data sets. The only outage is during the SWITCH phase, which will be much, much shorter than the BUILD2 phase outage.
As a result of the elimination of the BUILD2 phase, DB2 will need additional temporary storage for the shadow data sets for each NPI. Additionally, the cost of the REORG will increase because building an entire NPI will consume more CPU than the previous method of updating RIDs in the BUILD2 phase. Finally, you will need to modify any REORG jobs against different partitions that ran in parallel because the entire NPI will be built and then switched.
What about running REORG SHRLEVEL REFERENCE by partition? Well, REORG SHRLEVEL REFERENCE by partition will also rebuild any associated NPIs using shadows. So the same type of concerns apply (more disk storage, additional CPU, no separate jobs in parallel).
Reorganization and LOBs
Finally, as I blogged about before in the posting on LOBS, prior to V9 you could not access LOB data during a REORG. And a REORG did not reclaim physical space from the LOB data set because LOBs were moved within the existing LOB table space. V9 fixes these problems. During a REORG (in V9), the original LOB table space is drained of writers. All LOBs are then extracted from the original data set and inserted into a shadow data set. When this operation is complete, all access to the LOB table space is stopped (the readers are drained) while the original data set is switched with the shadow data set. At this point, full access to the new data set is enabled, and an inline copy is taken to ensure recoverability of data.