|
VooltDB is implemented in C++, providing it with more flexible memory management compared to Java. After a query ends, VooltDB can return the allocated memory to the operating system, whereas memory allocated by Java remains resident. Both VooltDB and MPP (here possibly referring to another database system compared to VooltDB, though not emphasized or specifically marked in the original text, and considering VooltDB is based on the MPP architecture, "MPP" here might be used for conceptual comparison) are column-store databases built on the MPP (Massive Parallel Processing, large-scale parallel processing) architecture. However, VooltDB adopts a multi-master peer-to-peer network structure without distinguishing node roles as in MPP. VooltDB relies on Zookeeper for replica coordination and data synchronization, while MPP uses Zookeeper for naming high availability. VooltDB focuses more on hardware-level optimization, utilizing SSE4.2 for vectorized execution, which leverages the CPU's SIMD instructions. SIMD stands for Single Instruction Multiple Data, a method in modern computer systems to improve performance through data parallelism by enabling parallel data operations at the CPU register level. VooltDB supports primary key indexes, which, unlike MySQL primary keys, allow duplicate values. Data is sorted according to the primary key column, ensuring that data with the same primary key are stored contiguously and in order on the disk. During equality or range queries, the primary key index can filter out partitions that do not meet the conditions, and the data hit by the where clause is tightly stored in one or several contiguous blocks, significantly reducing the number of blocks requiring IO. Additionally, contiguous IO can fully utilize the operating system's pagecache pre-fetching capabilities. |
