ClickHouse Enhanced Features

MPP Parallel Query

MPP parallel query: When a task is delivered to a cluster, parallel query and multi-copy parallel coordinated computing are performed on all nodes. Multi-core CPU parallel computing within a node and parallel computing between independent operators are performed as well.

Column-based Storage

Data of the same type is stored in a separate column. Only a few columns are obtained from multiple columns for analysis, which is efficient for analytical requests.

Vectorized Execution

Compared with the row-by-row processing mode in the traditional volcano model, the vectorized execution engine uses batch processing, which greatly reduces the function calling overhead, reduces the cache miss of instructions and data, and improves the CPU utilization.

Encoding and Compression

ClickHouse employs column-oriented storage. Data in the same column is continuously stored, and the underlying data is sorted during storage, which makes the regularity of data very strong and brings a higher data compression ratio. In some scenarios, the compression ratio of ClickHouse can reach 10:1, greatly reducing storage costs. In addition, the ultra-high compression ratio reduces the storage read overhead and enhances the system cache capability, thereby improving the query performance.

Multi-index Technology

ClickHouse supports partitions, primary keys, and secondary indexes. There are various types of secondary indexes. On the basis of column pruning optimization, table records are segmented horizontally. Indexes can be used to locate required block records, further improving query performance.

Materialized View

For high-frequency query modes, a small number of materialized views can be created to further improve query performance by consuming acceptable compute and storage resources.

Parent topic: ClickHouse

Previous topic: ClickHouse Application Scenarios

Next topic: Functions