Poor performance with Histogram

Hello listers.

I have used histograms to imporve performance drastically on several earlier occassions on our DSS databases. But recently, it took me quite a while to determine that by removing histograms, I am able to run one critical query - the report from which was happened to be of interest to my CEO - far faster than with histogram. The test query - involving two tables, one partitioned(60 Million) and another regular (13 Million), with outer joins, view merging, few aggregate sorts and sub queries- completes in under 5 sec without histograms while it takes nearly 17 minutes with histograms.

Can you tell me where I can look to understand this. Plans are different, of course. But how CBO fails to evaluate the plan I could not comprehend. I am in the process of studying Jonathan Lewis (Chapter 7 & 14) to understand. But any input/insight would greatly be appreciated.

Thanks,
Shiva

--
http://www.freelists.org/webpage/oracle-l