A few words on Index Compression
Posted by Harald van Breederode on April 5, 2010
For the last couple of months I have been quite busy delivering Oracle11g New Features courses and one of the topics covered is table compression. To introduce this topic I ask my students in which version the first compression feature was introduced. Strangely enough almost no one seems to know the correct answer, which is Oracle8i where index compression was introduced. Because table compression is very much like index compression I explain to my students what index compression is all about and quite often they are completely shocked by the fact that such an important feature is missing in their skill set. So it is about time that I post something about index compression.
Index compression works by eliminating duplicated column values in the index leaf blocks thereby allowing the storage of more index entries per leaf block. To demonstrate this feature I will start by creating a non-compressed index on a few columns on the CUSTOMERS table in the SH sample schema:
SQL> connect sh/sh Connected. SQL> create index i1 on customers(cust_first_name,cust_last_name); Index created. SQL> select leaf_blocks from user_indexes where index_name = 'I1'; LEAF_BLOCKS ----------- 193
As shown above the just created index consists of a total of 193 leaf blocks. To get an estimate about the percentage of leaf blocks that can be saved, by compressing the index, and the optimal compression factor we can use the information Oracle stores in the INDEX_STATS view after validating an index as shown below:
SQL> validate index i1; Index analyzed. SQL> select opt_cmpr_count,opt_cmpr_pctsave from index_stats; OPT_CMPR_COUNT OPT_CMPR_PCTSAVE -------------- ---------------- 2 47
The above output reveals that the optimal compression factor for this index is 2 and that this will most likely reduce the number of leaf blocks by 47%. So, let’s rebuild the index with compression enabled and verify the given estimate:
SQL> alter index i1 rebuild compress 2; Index altered. SQL> select leaf_blocks from user_indexes where index_name = 'I1'; LEAF_BLOCKS ----------- 101
The above shows that the number of leaf blocks in the index is now 101. We started with 193 leaf blocks and we were expecting a 47% reduction in the number of leaf blocks. 47% of 193 is 91 and 193 minus 91 is equal to 102. So Oracle estimated the reduction pretty good I would say ;-) Executing the validate index command again assures that 2 is indeed the optimal compression factor and no leaf blocks can be eliminated because the index is already compressed as shown below:
SQL> validate index i1; Index analyzed. SQL> select OPT_CMPR_COUNT,OPT_CMPR_PCTSAVE from index_stats; OPT_CMPR_COUNT OPT_CMPR_PCTSAVE -------------- ---------------- 2 0
Hopefully this little post is enough to trigger you to dive into the world of compressed indexes! For a much more thorough discussion on index compression see Richard Foote’s fabulous Oracle blog. (part 1, part 2, part 3 And part 4).