Thanks,
Charlie

Thanks for this, especially the tip about the types of updates that adversely affect bitmap indexes – this is exactly the sort of information I was trying to deduce with the tests I was performing; also thanks for the tip on leaf_blocks vs. blocks on user_segments.

I’ll be putting some more test cases together soon, so will follow your advice on putting the test data together – again, this will be very useful.

Many thanks, Mark.

The trick of using btree indexes in Star Transformations has been around since at least 9.2 – to paraphrase the book (p. 204) “there is no reason why the optimizer can’t decide to do a bitmap/btree conversion at any point in the execution plan”.

The thing about DML on bitmap indexed columns applies most significantly to large numbers of small changes. Adding 1M rows as a single change to 7M rows is likely to be the least damaging strategy – although I would have expected the leaf_block count to nearly double in at least some of the indexes. (A less ambiguous test of size would have been to check the leaf_blocks on user_indexes et. al. rather than blocks on user_segments). A strategy of adding (say) batches of 100 rows and committing, or even – as happens sometimes – single row commits could have a much worse effect prior to 10g. It’s very important to get a good model of how data chnages will actually occur. (Watch out, also, for big deletes followed by big inserts – nasty things can happen at random intervals for obscure internal technical reaons.

Data modelling for bitmap testing can easily be misleading – copying the same data set N times, for example, will produce a cyclic pattern in the data – hence the indexes – that can give some artificially beneficial, or detrimental, effects on some tests. I tend to use the mod(), trunc() and combined mod(trunc()) functions to produce deliberate patterns in the data as a way of approximating real data patterns.

Mr.Lewis’s first book is only for the “fundamentals” he mentions :) So I guess parallel sql, more partitioning, more bitmap indexes and conversions, query rewrite, star transformation etc. will be in the coming books.

I have got Jonathan’s CBO book at home with me, I did have a quick look through to see if the phenomenon was mentioned but I couldn’t see anything; if anyone has seen it in the book and knows the page number, I’d be most grateful.

– regards, Mark

More often than not, I have seen this b tree conversion in problem plans. But such is the nature of these things, that is probably because I’ve not needed to be aware of plans where this is not a problem.

Unsurprisingly, I’m pretty sure Jonathan Lewis covers this in his CBO book.