VoltDB概念

PARTITION: A partition contains a logical subset of each partitioned table's rows. For example, if Table T has rows (A, B, C, D), partition 1 contains the subset (A, C) and partition 2 contains the subset (B, D).

All VoltDB nodes in a cluster run identical software. All nodes are peers. There are not specific routing nodes, storage nodes, or computation nodes - they're all the same.

Each node has a TCP/IP connection to every other node. There is a fully connected socket mesh. Nodes communicate by sending messages over these sockets.

VoltDB is a Java program. However, SQL execution, table data storage, table indexes and views are implemented in C++ and accessed via JNI. All user data is stored on the native (non-Java-GC'd) heap.


Each node can be divided roughly into a top-half and a bottom-half. The top-half is responsible for I/O, inter and intra node communication, and the per-node aspects of the transaction ordering protocol. The bottom half is responsible for transactionally running stored procedures, executing SQL and storing user data.