Oracle BI Reports not hitting table partition

Question

Hi,
I have Fact Table 'F1' which is joined with Dim Table 'D1' on F1.Date1 = D1.Date1 Column. On Fact Table 'F1', we have partition on column "F1.Date1"
Partiton_Name Range
Jan10 - 01-Jan-2010 to 31-Jan-2010
Feb10 - 01-Feb-2010 to 28-Feb-2010
In Dim Table 'D1', I have Column "Month_Year" as well i.e., Month_Year ("Jan-2010", "Feb-2010").
Chart shows you data for multiple months (i.e., Jan-2010 - 1000000, Feb-2010 - 1500000) and I have navigate on column "Month_Year" to another report.
when i click on column "Month_Year" value i.e., Jan-2010 from report1 - Value "Jan-2010" goes a filter in another report (since i have given IS PROMPTED on second report).
Since I have partitioned on Column F1.Date, it goes for full scan (hitting all partitions exits in table 'F1'), I want to hit only 1 partion.
Please suggest.
Let me know in case of any clarification required.Thanks

Answer 1

I agree with your idea that partitioning for 40K rows (with the information you provided) seems excessive. Doing the math 40K rows * 218 partitions should come out to be only around 8M, which I would not immediately consider a candidate for partitioning.
There are benefits to partitioning that you're aware of; most of these involve the use of large data sets. Pruning is a huge benefit by excluding vast amounts of unwanted data from join consideration.
There are downsides too, mostly buried under the volume of data that partitions usually work with. There is a slight performance hit on standard queries against a partitioned table while SQL decides which partitions to hit. You can observe this by querying from the table, then by a specific partition when only trying to get data from a specific partition.
There is also the issue of how many partitions do you want to keep track of? You are asking good questions.
Its possible that the partitions are helping performance; the only way to make sure is to create a non-partitioned table using CTAS, index and constraint it like the partitioned table, and run some performance tests.