Oracle FAQ | Your Portal to the Oracle Knowledge Grid |
Home -> Community -> Usenet -> c.d.o.misc -> analyzing tables
I am developing a large DSS application and want to minimize the time spent in analyzing tables. I have some questions on the value/cost of analyzing tables.
(I will summarize all answers I get and send them to everyone who replies
to me)
One can analyze a table, an index or a column. This leads to collection of a set of statistics such a histogram for columns, number of datablocks occupied, how full they are etc.
(a) which statistics are the most useful to collect and which ones are the
most time consuming to collect.
For example, if a column is uniformly distributed, I should be able to do without building a histogram for it.
(b) Does not building a histogram lead to substantial savings in running
time?
(c) Does sampling only apply to collection of histograms or even to the
other statistics that are collected i.e. if data is uniform, can I use a low sample percentage and get all statistics (other than the histograms) collected cheaply and accurately.
(d) What sampling percentage is reasonable when data is skewed.
I am really intersted in hearing about what worked in practice rather than claims based on statistical theory (I suspect that assumptions needed by the theory are not true).
thanks,
-Waqar Hasan
hasan_at_cs.stanford.edu
Received on Mon Dec 22 1997 - 00:00:00 CST