Skip to main content

Understanding database [8] : Bitmap indexes

Understanding database 8 : Bitmap indexes

The idea of a bitmap indexes is to record values of sparse columns as a sequence of bits, where one bit would represent for each possible value. The bit for the row, would indicate whether if it has this value or not. Formally, we may consider it as a collection of n bit-vectors, where we have one for each possible value f.

Six vector operations are supported by bitmap indexes, namely COUNT, LENGTH,OR,AND, NOT and XOR. In order to find the correct bitmap vector for a given value ,we may build a B+ tree over the possible values, with the leaves storing the index entries. The modification of bitmap index based data needs to be dealt with caution : the record numbers must remain fixed once assigned, and the changes to the data file require the bitmap to change as well by setting or clearing corresponding bits in vectors. If a new record is to introduce a new value, we may have to introduce a new bitmap vector for that value.

As we may derive, we will need O(mn) space for the bitmap index, while m stands for the number of distinct values for field f, and n stands for the number of records. The problem is that the more distinct values we have, the sparser the bitmap vectors will be, and we will end up storing heaps of 0s in the vectors. For that reason, it would be a good idea to compress bitmap vectors for efficient storage.


Popular posts from this blog

Understanding database [9] : Choosing indexes

Understanding database 9 : Choosing indexes When choosing indexes, we choose the best plan that suits for the queries, and look for additional indexes that may potentially upgrade upon that. Before creating, we must also consider the impact on updates in the workload, such that indexes take disk space. For a query, the WHERE clause are the main focus point to make indexes on, where exact matches suggest a hash index and range queries suggest a tree index. Clustering is extremely helpful when it comes to range queries, and may also help with equality queries if there are duplicates. Search keys with multiple attribute should be considered if a WHERE clause contains multiple conditions, and the order of attributes is important for range queries. Searching may become ‘index-only’ with such indexes.

Understanding database [6] : Clustered Index

A clustered index is what is good for a range search over a range of search key values, and the index entries and the rows are ordered the same way, which is different from unclustered indexes (secondary indexes). To use a clustered index, we use the index to locate the first index entry at the start of the range, which where the first row is located at. If the index is clustered, subsequent rows will be stored in successive locations with the ordering , therefore it may minimize the page transfers and maximizes cache hits. For one table, there may only be one clustered index, whereas there can be as many unclustered indexes as you may want  to create. Unclustered indexes aren ’ t as good as clustered, but they might become necessary when it comes to finding for other attributes apart from the primary key. Smaller topics : Dense index & Sparse index When we say sparse index, we mean that there is an index entry for each page of the data file. With this structuring