A typical T24 table consists only of two columns, a RECID (as primary key) and an XMLRECORD. This is similar to other systems implemented based on XML data, both OLTP and more of OLAP/BI character.
Inlining of dataFor these kind of tables it is a good idea to INLINE the XML column. Inlining is good for several reasons. Because XML data is regularly stored in a special XML Data Area (XDA), its access requires an indirection from the row data. The descriptor that is stored in the row is taken as input to look up the actual storage location via the XML Regions Index. Eliminating the descriptor means the XML data is directly fetched with the row as well as fewer entries in the XML Regions Index.
If there is test data, the function ADMIN_EST_INLINE_LENGTH can be used to find the typical inline length that should be set. The maximum inline length depends on the page size and hence the maximum row size. Both are documented with the just mentioned function. For T24 the recommendation is to use an inline length of 32000 bytes or even more in 32 kb pages.
Volatile Tables for Index UseAnother configuration option for tables is to declare them as VOLATILE. The background is explained on this page with "Relational index performance tips". Basically, the optimizer will prefer index-based access to the table even if the cardinality changes frequently.
Enable static compressionMany systems benefit from data compression. The same goes for T24 on DB2. However, the question is whether to use static (table) compression or even the newer adaptive compression (page-level compression on top). Based on different benchmarks and experience with production-oriented testing the recommendation is to use static compression. The additional CPU overhead for a higher degree of space and IO savings impacts system throughput. When in doubt test it on your own. Another effect: Indexes on compressed tables will also be compressed by default.
Consider APPEND ON or DB2MAXFSCRSEARCH=1With APPEND ON as an option to CREATE/ALTER TABLE, DB2 will insert "at the end" of the table without searching for free space elsewhere. It results in slightly higher space consumption, but benefits insert performance. An alternative is to set DB2MAXFSCRSEARCH to a low value, e.g. to 1. It determines on a global level of how many free space control records (FSCR) to search for placing a new record. Thus, all tables would be impacted, not just the one where APPEND ON is specified.
Separate tablespaces for table, index, and long dataIt is always a good idea to have separate tablespaces for the different page types. They can be specified during CREATE TABLE.
PCTFREE 99 for hotspot tablesDepending on the application and usage type of T24 there can be hotspot tables. For these it could make sense to specify a high value for PCTFREE during CREATE/ALTER TABLE. This determines how much space is left free (read: unused) in a page, i.e., how many or how few records are stored in a single page. The fewer records - the extreme would be a single record - in a page, the less likely that page becomes a hotspot.
Consider table/range partitioning for bigger tablesLast, but not least, it is a good idea to apply range partitioning on bigger tables. This benefits performance in several ways: Old data can be rolled out (DETACHed) quickly, queries perform faster due to range elimination and rebalancing of work, and maintenance can also be done in parallel on the ranges.
The partitioning can either be done on the RECID (the primary key) or a new hidden column could be introduced to evenly split the data into ranges, depending on needs.
SummaryTuning basic properties for XML-based tables is not (that) different from other tables. The above gives an introduction of should be considered when setting up Temenos T24 or similar XML-based systems on DB2. I didn't provide examples with the full syntax. If you need them, leave a comment... ;-)
Special thanks to Juergen Goetz for sharing his experience for use with other customers and this blog entry.
BTW: There are also few links for Temenos T24 on DB2 on my resources page. General DB2 system setup for HA for Temenos is discussed in the Redpaper "High Availability and Disaster Recovery for Temenos T24 with DB2 and AIX"