Re: analyze indexes

On Sat, 13 May 2000, Rajagopal Venkataramany wrote:
> My observations on the situation :
>
> 1. If you do not drop the indexes before the load process, if the load
> volume is high, you might end up spending more time loading the
> data. Generally, it is not a recommended practice.

Unfortunately, it may have to be done. Currently, a direct-path load takes 5 minutes, but rebuilding the indexes takes 4 hours. Testing with the same amount of data (loading ~3.5 million rows) takes about an hour total with the indexes left on. Thus, we save 3 hours right off the top.
I would like to move back to a direct load once we figure out the exact problem that is occuring with the indexes.

> ** When you rebuild the index, the storage requirements would be high
> as the existing index space is still retained till the rebuild
> is completed. In other words you would require twice the size of the
> index space when you attempt to rebuild. Alternatively, you can
> drop and recreate the index details, if that is feasible.

Which is what is occuring now - we drop the indexes, load the data, then recreate the indexes. Unfortunately, with the amount of data currently in the table, creating unique and non-unique indexes takes quite some time.

Thanks,
GC

--
There is something inherently wrong with the world when Bill Gates is a
famous billionaire and Dennis Ritchie lives in relative obscurity.