OraNA :: Oracle News Aggregator

So, which number takes more bytes inside an Oracle row?

A: 123

B:  1000000000000000000000000000000000000

And the correct answer is … (drumroll) … A! The “big” number 1000000000000000000000000000000000000 actually takes less space than the “small” 123!

Let’s verify this:

SQL> select vsize(123) A, vsize(1000000000000000000000000000000000000) B from dual;

         A          B
---------- ----------
         3          2

WTF? Why does such a small number 123 take more space than  1000000000000000000000000000000000000 ?

Well, the answer lies in how Oracle stores numbers. Oracle NUMBER datatype doesn’t store numbers in their platform-native integer format. Oracle uses it’s own format which stores numbers in scientific notation, in exponent-mantissa form. More details about this here.

You can use the DUMP sql function to see the actual binary value of the number data stored:

select dump(123) from dual;

DUMP(123)
---------------------
Typ=2 Len=3: 194,2,24

SQL> select dump(1000000000000000000000000000000000000) from dual;

DUMP(10000000000000
-------------------
Typ=2 Len=2: Typ=2 Len=2: 211,2

So, although the number 1000000000000000000000000000000000000 is bigger than 123, when stored in base-10 exponent form, it really carries much less information in it than 123 (1 x 10^36 vs 123 x 10^0). Oracle doesn’t need many bits for keeping the precision of this large value as it happens to be a power of 10.

See what happens when I store a number only slightly bigger or smaller than the original large number, now the stored number requires much more storage for keeping the required precision:

SQL> select dump(1000000000000000000000000000000000000+1) from dual;

DUMP(1000000000000000000000000000000000000+1)
-------------------------------------------------------
Typ=2 Len=20: 211,2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,2

SQL> select dump(1000000000000000000000000000000000000-1) from dual;

DUMP(1000000000000000000000000000000000000-1)
-----------------------------------------------------------------------------------------
Typ=2 Len=19: 210,100,100,100,100,100,100,100,100,100,100,100,100,100,100,100,100,100,100

Here’s the list of events where I’ll speak this year:

Michigan OakTable Symposium 2010
Ann Arbor, MI
16-17 September 2010

Considering the concentration of OakTable members there, this will be an awesome event!

I will be delivering my “Back to Basics: Choosing The Entry Point to Performance Troubleshooting Wisely” and “Understanding LGWR, log file sync waits and commit performance” sessions there.

Promo video:
http://www.oaktable.net/media/michigan-oaktable-symposium-2010-promo

Agenda & Registration:
http://michigan.oaktable.net/

Oracle Open Closed World
San Francisco, CA
19-22. September

Note that I won’t be speaking at the official Oracle Open World conference, but I will be speaking at a secret underground event there, about some really fun stuff, like deep internals, hacking, kernel tracing and of course advanced troubleshooting ;-) And rest of the time I’ll be in some bar.

NYOUG Fall 2010 Training Session
Manhattan, NYC, NY
16 November 2010

This is a full day seminar organized by NYOUG. I will be delivering my “Scripts and Tools for Oracle Troubleshooting and Advanced Performance Analysis” session there. It’s an updated version of the material I delivered at the Hotsos Symposium Training Day this year.

Agenda & Registration:
http://www.nyoug.org/upcoming_events.htm#NYOUG_Training_Days

UKOUG Tech & EBS Conference (to be confirmed)
Birmingham, UK
29 November – 1 December 2010

I submitted four papers to UKOUG Tech&EBS conference, so if all goes well, I’ll be there in end of Nov/beginning of Dec too.

http://techandebs.ukoug.org/