Parallel Execution

Inter-Operator (Vertical) Bushy

For transactions within terabytes of data, the parallelism is mainly in transaction level. Transactions and queries involving more than terabytes of data will be divided into sub tasks and executed in independent disk servers.

Joins

Nested Loop Join Hash Join Sort-Merge Join Semi Join

NonStop SQL supports hash join, nested loop join, and sort-merge join. The developers of NonStop SQL didn't devote lots of effort on distributed join optimization, because they thought those optimization makes little sense. To the best of our knowledge, NonStop SQL supports semi-join.

Storage Model

N-ary Storage Model (Row/Record)

NonStop SQL uses the N-ary Storage Model, which is appropriate for heavy OLTP workloads.

Query Interface

SQL

NonStop SQL adopted the SQL interface because SQL supports views, assertions, and integrated dictionary. Those were what customers asked for by that time.

Query Compilation

Code Generation

The query compiler transforms the SQL source commands into executable query plans in host language. The executable query plans and the SQL source commands are stored in a newly created object program file.

Storage Architecture

Disk-oriented

The data of every table is stored in one or more disk files. Each disk file has a label storing the name and catalog information of the associated table.

Indexes

B+Tree

NonStop SQL supports B+ tree index. By default, the primary key is used as the index of a table. To speed up the non-primary key look up, an index table can be created based on selected columns.

Data Model

Relational

NonStop SQL supports the basic features of relational database management systems. What distinguish it from many other relational DBMS are the additional features providing excellent OLTP performance, availability, and scalability.

Isolation Levels

Read Uncommitted Read Committed Repeatable Read

NonStop SQL supports read uncommitted, read committed, and repeatable reads. Users are able to select isolation levels by themselves.

System Architecture

Shared-Nothing Shared-Disk

NonStop SQL supports scale-out, and each unit is called a disk volume. Among the disk volumes, the distributed architecture is shared-nothing. In every disk volume, there can be multiple disk servers which are doing data managing, logging, locking, and query execution. Among the disk servers, the distributed architecture is shared-disk.

Query Execution

Tuple-at-a-Time Model

Views

Virtual Views

Whenever a view is created, NonStop SQL stores a view definition instead of the materialized table. Whenever a user selects data from a view, the data is directly retrieved from the underlying base tables.

Concurrency Control

Two-Phase Locking (Deadlock Prevention)

The users can set the lock granularity, lock protocol, and lock waits styles in the CONTROL TABLE. The deadlock prevention is based on lock timeouts.

Logging

Logical Logging

Every single transaction has one single corresponding log record. This log record contains information about how to perform redo, undo, and media recovery.