ROLLOVER STRATEGIES IN A N-BIT DICTIONARY COMPRESSED COLUMN STORE

摘要

Disclosed herein are system, method, and computer program product embodiments for rollover strategies in an n-bit dictionary compressed column store. An embodiment operates by receiving a new value for addition to a compressed column store, determining that a current memory block of a most recently added token to the compressed column store is the insertion block. It is determined that the maximum token value has been reached for the current memory block. A new virtual memory block is created using the current insertion block, and a token corresponding to the new value is stored in the new virtual memory block. In another embodiment, when it is determined a maximum number of token values that may be stored in a compressed column store has been reached for a data dictionary, the compressed column store is converted into a composite store include a flat store where the new value is stored.

主权项

1. A method, comprising:
receiving a new value for addition to a compressed column store, the compressed column store including a plurality of tokens, each token corresponding to a value in a data dictionary, and being associated with a row identifier (RID) in the compressed column store; determining an insertion block of the compressed column store, the insertion block being a physical or virtual memory block where new tokens are inserted; determining that a current memory block of a most recently added token to the compressed column store is the insertion block, the compressed column store including one or more memory blocks each with a maximum token value that may be stored within a respective memory block based on an encoding of the respective memory block; determining that the maximum token value has been reached for the current memory block based on type of encoding of the current memory block; creating a new virtual memory block using the current memory block, wherein the new virtual memory block has an encoding greater than the encoding of the current memory block, and wherein the new virtual memory is designated as the insertion block; and storing a token corresponding to the new value in the new virtual memory block, wherein the token corresponding to the new value in the new virtual memory block is accessed in a same manner as existing tokens in the current memory block.