Re: Index advisor and system cross-reference files

On 30-Aug-2015 21:35 -0600, satya wrote:

Can someone please confirm the following advise from Index advisor
is normal for iseries and not a problem specific to our system? I am
finding it difficult to believe that system tables do not have
proper indexes (Index Advisor point of view).

That there should be some particular /proper indexes/ is surely apocryphal; what the system ships, is what is at least somewhat generally acceptable, but _necessarily_ the variety provided can not be comprehensive across every customer system, because the [cardinality of the] data will be specific to each system... and the storage and maintenance impacts\requirements also necessarily limit the number of indexes that could be reasonably provided as part of the OS.

I found couple of suggestions by Index Advisor on queries using
views (QSYS2.SYSTABLES, SYSIBM.SYSCOLUMNS, etc.. ) related to system
cross-reference files on our test LPAR. This LPAR was recently
upgraded to V7R1 from V5R4 by IBM team. There is no business
application activity on this LPAR at present.

One of the queries with index advised is "SELECT SYSTEM_TABLE_NAME
FROM QSYS2.SYSTABLES WHERE TABLE_SCHEMA = ? AND TABLE_NAME = ? FOR
FETCH ONLY WITH NC". The query is executed by QDIRSRV. Index advised
is on QADBXREF (columns DBXREL , DBXLB2 , DBXLFI, DBXATR).

The query should be served quite well by the keyed Access Path (ACCPTH) provided by the logical file QADBXLLB in QSYS [keys: (DBXLB2, DBXLFI)]. Despite an advised non-existent AccPth, that file QADBXLLB was probably identified as the chosen AccPth for the implementation? Verify the Page Size (PAGESIZE) is as large as the SQL Query Engine (SQE) prefers... though I am not sure if there is a documented procedure to /correct/ that [I recall having an internal document, restricted because the negative impacts and required recovery], if that size attribute is the older-fashion Key-Length (*KEYLEN) setting.

The specifically advised index could be created [by a user with the *ALLOBJ Special Authority; the authority given to *PUBLIC is immaterial], but the query is unlikely to perform much [if any] better than implementing with the Access Path for QADBXLLB.