Composite Group-Keys - Space-Efficient Indexing of Multiple Columns for Compressed In-Memory Column Stores

Real world applications make heavy use of composite keys to reference entities. Indices over multiple columns are therefore mandatory to achieve response time goals of applications. We describe and evaluate the Composite Group-Key Index for fast tuple retrieval via composite keys from the compressed partition of in-memory column-stores with a main/delta architecture. Composite Group-Keys work directly on the dictionary-encoded columns. Multiple values are encoded in a native integer and extended by an inverted index. The proposed index offers similar lookup performance as alternative approaches, but reduces the storage requirements significantly. For our analyzed dataset of an enterprise application the index can reduce the storage footprint compared to B+Trees by 70 percent. We give a detailed study of the lookup performance for a variable number of attributes and show that the index can be created efficiently by working directly on the dictionary-compressed data.