Understanding DB2 UDB for z/OS Administration
This section discusses:
The section discusses:
Tablespaces named xxLARGE—where xx is a product identifier, such as HR—contain tables that grow substantially and/or experience high update activity. You should track the growth and extents for tablespaces and indexes, as well as monitor for page splits in the indexes.
Each of the tables in xxLARGE is a candidate for partitioning or for a separate tablespace. Tables defined in tablespaces other than xxLARGE are relatively stable and can be defined in shared tablespaces with little, if any, freespace.
As a general rule of thumb, the xxLARGE tablespaces grow substantially large with application data and contain the largest tables in your database. From a PeopleTools perspective, there are several delivered tablespaces that may grow in size. For example, tablespace PTTLRG contains PeopleTools tables (XLATTABLE, PSPCMNAME, and others) that my grow large in size. The “Tree” tables are delivered in tablespace PTTREE—tables prefixed with PSTREE%. These tables may grow substantially early on as you add branches and nodes in the Tree Manager, then plateau once the tree structure is fully defined.
Customers with large amounts of data may require that the larger tables be partitioned, and as a result must be moved to their own tablespace. This improves concurrency and also allow DB2 UDB utilities such as backup, reorg, and Runstats to be run in parallel.
With PeopleSoft 8, new tablespaces were introduced for tables requiring row level locking to avoid deadlock and timeout errors. Those tablespaces are: PTLOCK, PTAMSG, PTPRC, PTRPTS, PTAUDIT, PTPRJWK, PTCMSTAR and PSIMGR. Note that PSIMGR and PSIMAGE both require a 32K page size. If redistributing any of the tables delivered in these tablespaces, it is critical for performance to carry over the row level locking attribute and buffer pool assignment for the new tablespace.
You can avoid reaching lock escalation thresholds by ALTERing tablespaces from LOCKSIZE ANY (or PAGE) to LOCKSIZE TABLESPACE for the duration of batch jobs. This technique also improves batch program performance.
The ALTERed LOCKSIZE specification is effective immediately. Plan rebinds are not needed since PeopleSoft uses dynamic SQL. The simplest way to implement this technique is to ALTER all of the application tablespaces. If that is not desirable, determine the tables accessed in a particular job by examining SQL statements in PS_SQLSTMT_TBL and finding their corresponding tablespaces.
Note: Tablespaces should be ALTERed back to the original value after job completion. Tablespace locks will lock out online users until LOCKSIZE is reset to PAGE or ANY. If online users are active during the time you are running batch jobs, you may not want to ALTER LOCKSIZE to TABLESPACE.
This section discusses:
Isolation levels and CURRENTDATA
For PeopleTools, the use of Cursor With Hold (persistent cursors) with PeopleSoft applications is controlled entirely by PeopleTools. Consequently, there is no reason to use anything other than the IBM default for CURSORHOLD.
Isolation Levels and CURRENTDATA
PeopleSoft batch processes interface to DB2 UDB either through PTPSQLRT (for Cobol and AE), or through SQRPLAN for SQRs. Both of these are bound with the defaults—that is, CS (cursor stability) and CURRENTDATA NO. Using CURRENTDATA NO results in less lock contention in DB2 UDB and potentially reduce deadlock situations. It also provides two extra benefits:
Block fetch is enabled for ambiguous cursors.
DB2 UDB considers parallelism for ambiguous cursors.
A DB2 z/OS subsystem parameter RELCURHL lets you indicate that you want DB2 UDB to release a data page or row lock after a COMMIT is issued for cursors defined WITH HOLD. This lock is not necessary for maintaining cursor position.
The default for DB2 z/OS is YES. In prior releases, the value was NO, which causes DB2 UDB to hold a data page or row lock for the row on which the cursor is positioned. This lock is not necessary for maintaining cursor position and could cause deadlocks. The PeopleSoft recommendation is Yes to improve concurrency.