Re: Varying fields and SQL predicate LIKE

On 22-Jul-2010 08:12, Charles Wilt wrote:

I've got a 3.1 million row table with a 25 character field that
holds a street address, ex. '123 MAIN ST.'

We currently perform seraches by address like so:
select * from myTable where address like '%MAIN%'

Given the leading '%', I know the search will require a full
table (perhaps full index) scan.

Given that the average TRIM'd length of the data in the field is
16 and that RPG works much faster with varying fields. I was
thinking that if I changed the field to varying, the DB wouldn't
have any trailing blanks to compare too and thus my searches
should be about 36% faster...

However, I'm not seeing a 36% improvement. In fact, it appears
that performance is worse by a couple of percent!

Any thoughts as to why this is the case?

I've double checked the following:
- The varying field is set ALLOCATE(25)
- The data in the varying field is TRIM'd

Since a blank is not the prefix of the data being searched, you could not see the maximum potential improvement you are looking for. Try searching on '% MAIN%' instead, to effect a better test; not that I trust the performance test results for both non-dedicated and without averaged repeated tests including reversed order of any two compared scenarios.

By the way, also try ALLOCATE(0) [or whatever syntax has that effect] to force all of the data into its own segment. Depending on optimization, at the LIC DB level that storage could be the best possible means to reduce the I\O; the I\O is the greatest impediment for completing the processing the quickest. If the query has the capability to interrogate only the AuxSID, that could be similar to an index-only access, whereby all of the processed data is effectively contiguous.

Regards, Chuck