Comments on: Row Directory

By: Rowid « Oracle Scratchpad

Rowid « Oracle Scratchpad — Sun, 09 May 2010 19:48:15 +0000

[...] is to tell me which entry in the block’s “row directory” I occupy, and since the row directory can hold 4,096 entries that takes two bytes. So – from my perspective – the cost of my [...]

By: Row count 2 « Oracle Scratchpad

Row count 2 « Oracle Scratchpad — Wed, 05 May 2010 19:43:54 +0000

[...] first extract is from the block dump of the first block after the update and shows the row directory and the start of each row in the “row heap” (that’s a term I’ve just [...]

By: Row count « Oracle Scratchpad

Row count « Oracle Scratchpad — Sat, 17 Apr 2010 11:13:21 +0000

[...] other “magic number” of 4,096 is there because that’s the maximum number of row directory entries you can have in a single [...]

By: heap block compress « Oracle Scratchpad

heap block compress « Oracle Scratchpad — Tue, 30 Mar 2010 19:25:45 +0000

[...] in case a process rolled back the delete. (This tied back to a note I had written showing that the row directory in a table block could become much larger than you might think [...]

By: Index too big « Oracle Scratchpad

Index too big « Oracle Scratchpad — Fri, 26 Mar 2010 22:42:09 +0000

[...] until after the commit; however tables and indexes are handled differently. Tables only need to keep a “stub” to represent the row that has been marked for deletion, but indexes have to keep the entire index [...]

By: Pseudo-parallel « Oracle Scratchpad

Pseudo-parallel « Oracle Scratchpad — Sun, 03 Jan 2010 11:52:13 +0000

[...] 4095 as the “highest row in block” value to calculate the high rowid. (There’s an interesting consequence of this hard limit that could cause surprises in data warehouses and other system (e.g. Peoplesoft) [...]

By: Lunchtime Quiz « Oracle Scratchpad

Lunchtime Quiz « Oracle Scratchpad — Sat, 26 Sep 2009 03:26:27 +0000

[...] The answer is here, but do read the comments on this post before you look at it. Comments [...]

By: Glossary « Oracle Scratchpad

Glossary « Oracle Scratchpad — Sat, 12 Sep 2009 19:10:01 +0000

[...] Some symbolic block dumps (table and undo particularly) will list the row directory, others (e.g. index) will not. There are variations on a theme – blocks from heap tables and clusters also contain a “table directory”, but this will contain only one entry in the case of a simple heap table; blocks from a compressed index include a “prefix directory”; blocks from a compressed table include a “token directory”. In all cases the concept is the same: a list of two-byte entries pointing to a piece of useful data. (See also: this blog entry) [...]

By: Jonathan Lewis

Jonathan Lewis — Mon, 20 Jul 2009 12:46:28 +0000

You need to do some simple testing. Create a table with a single column of varchar2(12), say, insert 1,000 rows with the same value and dump a block; then compress the table, dump a block, and compare the two dumps.

Here’s an extract from a dump of such a compressed table:

tab 1, row 702, @0x11be tl: 5 fb: --H-FL-- lb: 0x0 cc: 1 col 0: [12] 78 78 78 78 78 78 78 78 78 78 78 78 bindmp: 2c 00 01 01 00 tab 1, row 703, @0x11b9 tl: 5 fb: --H-FL-- lb: 0x0 cc: 1 col 0: [12] 78 78 78 78 78 78 78 78 78 78 78 78 bindmp: 2c 00 01 01 00 tab 1, row 704, @0x11b4 tl: 5 fb: --H-FL-- lb: 0x0 cc: 1 col 0: [12] 78 78 78 78 78 78 78 78 78 78 78 78 bindmp: 2c 00 01 01 00

Note that the row lengths are 5 bytes (tl: 5), or substract two adjacent addresses.Since there are three bytes of row overhead (flags, lock byte, column count) that means only two bytes have been used for the column value – so possibly one byte to flag that the next byte is a token, and one byte for the “token pointer”.

You stil have to do a few more tests, of course but at first sight it looks like a repetitive four-byte string can give you some space benefit on compression.

Further tests: if a “token pointer” is one byte does that mean a block can hold only 255 different token values ? Maybe that one byte can turn to two bytes.

What happens if two adjacent columns have very repetitive combinations – you get one token to represent two columns, does that change anything else.

By: Pramode

Pramode — Fri, 17 Jul 2009 19:37:46 +0000

Does any one know in Oracle 10g when table is compressed and in a block there are duplicate values for one column what is the size of the reference pointer size?
Ex: if i have column value “test” (4 bytes ) repeated for the whole block. what is the size if the pointer?

My concern is if the pointer size is 4bytes or 8 bytes then compression may not be helping me..

Thanks
Pramode