dbms_stats vs analyze table

This was an interview question asked:

What is the major difference between analyze table versus dbms_stats.gather table stats

I answered Oracle strongly recommends using dbms_stats for gathering statistics.

Scenario :
Suppose there are 100 million of records getting inserted in a table, and there is an index on a column. What would be your option for performance tuning, would it be analyzing table stats ... rebuild index or dbms_stats. gather table stats with Cascade = True..

Which is better, why?

Since there is an index on table, which one would actually be slower. Would setting cascade=True slow down the performance.

My answer: I said using dbms_stats.gather_table_stats with setting Cascade= True would be better.

However, I am not sure about this. Could someone answer..

Quick response is appreciated
Who is Participating?

[Webinar] Streamline your web hosting managementRegister Today

DavidConnect With a Mentor Senior Oracle Database AdministratorCommented:
Thanks for the follow-up.  I happened to learn something this afternoon about the CBO that I should have known -- and will try to use the next time I interview someone :)

I tend to get too literal about some things, so this may sound like I've missed the point, again.  But your answer is certainly affected by which RDBMS version is in question, as the command syntax changes.  ANALYZE in 11g is not used for the data dictionary (tables), but rather to validate the index structure.  According to my source, however, cascade may be either COMPLETE or FAST, but your example "TRUE" is illegal syntax.  

If you're thinking of DBMS_STATS, then the correct syntax seems to be ... no_invalidate=TRUE.  Again from the docs, "This parameter specifies when dependent cursors cached in the library cache will get invalidated when statistics are gathered on referenced tables or indexes....specify when to invalidate dependent cursors i.e. cursors cached in the library cache area of the shared pool which reference a table, index, column or fixed object whose statistics are modified by the procedure call."

Now, I'm likely to be doing this all wrong, but all the versions between 7 and 12 can run together in my memory.  Usually I'd simply disable indexing from the table(s) during such a data load, then rebuild the index(es) on completion.

If you cannot disable or drop the existing indexes, you might be loading these rows in archivelog mode, meaning you're generating boo-koodles of redo and archive logs.  

Lastly, which type of indexes, and their skew, are factors.  I hope this helps more than it confuses.
DavidSenior Oracle Database AdministratorCommented:
In 10g, ANALYZE offers value in data collection not related to the cost based optimizer (CBO), and DBMS_STATS is encouraged.  In 11g, ANALYZE is deprecated.

For another source see http://asktom.oracle.com/pls/asktom/f?p=100:11:0::::P11_QUESTION_ID:1632603200346280697.
d27m11yAuthor Commented:
Can you please answer my question when Indexes exist on a table
Never miss a deadline with monday.com

The revolutionary project management tool is here!   Plan visually with a single glance and make sure your projects get done.

d27m11yAuthor Commented:
If I understand correct, analyze index syntax can be used even after 11g to validate the index structure and to my question, you suggest to drop all indexes and rebuild them..

So what should be my steps..

1. Drop all indexes
2. Perform dml operation
3. Rebuild indexes

Do I have to use analyze table compute statistics rebuild indexes  here.. am still confused once I follow the above steps.

Well,  to be clear when should I use analyze table compute stats rebuild indexes ..

Am sorry, it is quite confusing for me..
DavidSenior Oracle Database AdministratorCommented:
I think we're in synch.  To be clear about the drop, however, my experience has been to drop the primary INDEX of a table.  The remaining indexes, constraints and any triggers may simply be disabled.  There's some chatter about this but it involves other Oracle products and I don't want to throw you.

However, you haven't yet defined how your site inserts a hundred million rows into a table.  For example, is there content to which the new rows are added?  Or is the existing data truncated and laid fresh each time?  Is there partitioning?  Are you running from a command prompt, or Grid Control?  Are other users locked out during this process?  See where I'm going?  Lots and lots of variables.

The function I'll end up directing you to, is dbms_index_utl.build_table_indexes.  That's getting into a new topic, however, and I want to ensure this one is resolved first.
d27m11yAuthor Commented:
Helps me to resolve the issue
DavidSenior Oracle Database AdministratorCommented:
Thanks for the closure, and a good question.  Since your "B" grade indicates there's room for improvement, what caused the downgrade please.

All Courses

From novice to tech pro — start learning today.