Storage Model

Hybrid

Although the original Ingres was row-based, Actian has incorporated columnar storage into its latest version of Ingres (Actian X) to improve its performance on OLAP tasks.

Checkpoints

Consistent Blocking

Ingres supports two types of checkpointing: online and offline. Offline checkpointing requires all users to disconnect from the database. It acquires a global lock on the entire database and performs backup on all the data. Online checkpointing does not require users to disconnect and is the default checkpointing scheme. Therefore, it is faster than offline checkpointing as it results in less down-time. However, it is still blocking. Users cannot proceed until the Ingres creates a consistent point of the database. The DBMS records all changes in the database but does not support operations such as drop or create during online checkpointing.

Joins

Nested Loop Join Hash Join Sort-Merge Join

Ingres supports joins with hash join, sort-merge join, and nested loop join algorithms.

Compression

Null Suppression

Ingress's compression of tables only work on character and text columns. Trailing nulls and blanks are compressed. Non-nullable types such as Integer, floating point, and date are not compressed. It is not mentioned in the documentation that Ingres supports other types of compression schemes. Compression is applied to each page independently. It is shown in both the documentation and the original paper that compression in Ingres is useful in storing variable length domains. For example, if there is a column storing customer's reviews on a local business, then each review is padded at the end with nulls if it is shorter than the maximum allowed length to maintain homogeneity across the column. By applying compression on this column, these trailing nulls are compressed and the reviews take much less disk space because most of them are expected to be shorter than the maximum allowed length. However, the documentation suggests that compression should be used situationally, as decompression consumes CPU resources. It is up to the application to weight the trade-off between CPU utilization and disk space.

Concurrency Control

Multi-version Concurrency Control (MVCC) Deterministic Concurrency Control Two-Phase Locking (Deadlock Detection)

Ingres has five lock levels, which are, from most to least granular, Row, MVCC, Page, Table, Database.

Ingres has four types of locks: physical locks, logical locks, control locks, and value locks. A logical lock is acquired when a transaction starts. It is held while the transaction is alive and ends at transaction commit or abort. A physical lock is used within a transaction to protect resources. A control lock is a physical lock on the Table Level that protects a table while its schema is being modified to ensure that no transactions can access the table while such a modification is in progress. A value lock is a physical lock on the Row or MVCC level. It is used when the isolation level is set to serializable to avoid phantom reads, which can occur under ranged queries if this type of locks is not present.

Ingres supports the following lock modes: NL(no lock), IS(intent to share), IX(intent to update), S(shared), SIX(shared with intent to update), and X(exclusive access).

The locks in Ingres are managed by a lock manager. They are tracked using their unique lock IDs. The DBMS implicitly requests, escalates, and releases locks based on the statement or command, but it also provides commands for explicit lock management.

It supports deadlock detection and aborts one of the transactions that are deadlocked. The aborted transaction is rolled back and the user is notified of an error on the aborted transaction.

Storage Organization

Heaps Indexed Sequential Access Method (ISAM)

Heaps and ISAM are supported by Ingres.

Foreign Keys

Supported

After primary keys are created, a user can create foreign keys using the New Foreign Key Dialog.

Isolation Levels

Read Uncommitted Read Committed Serializable Repeatable Read

Ingres supports four isolation levels, from favoring consistency to maximizing concurrency: Serializable, Repeatable Read, Read Committed. and Read Uncommitted, Serializable is the default isolation level and it provides the strongest consistency guarantee.

Query Execution

Materialized Model

Ingres executes queries by producing materialized views.

Data Model

Relational

Ingres is a relational database system.

Query Interface

Custom API SQL Stored Procedures Command-line / Shell QUEL

Ingres was originally built on UNIX, hence it supports a command-line interface. It also supports SQL and QUEL and stored procedures in these two query languages. In addition, applications can use OpenAPI, a set of APIs in the C programming language, to interact with Ingres.

Logging

Physical Logging

Ingres uses physical logging, keeping BEFORE and AFTER images for database updates. It chooses to record these images, not the updates, to make sure that all logs are idempotent. Ingres optimizes BEFORE images to save disk space such that if multiple records on the same page are to be updated, then only one BEFORE image is generated.

Views

Materialized Views

Ingres produces materialized views.

Query Compilation

Stored Procedure Compilation

Ingres allows users to write stored procedures in one of the supported query languages.

Stored Procedures

Supported

Ingres supports stored procedures in both SQL and QUEL.

System Architecture

Shared-Everything

Ingres is a single-node database management system, and therefore it is "Share-Everything".

Indexes

B+Tree Hash Table R-Tree

A user can create an index on an existing table in Ingres using the CREATE INDEX command. An index cannot contain more than 32 columns, but one can create as many indexes as needed on a table. Ingres chooses ISAM (Index Sequential Access Method) as the index data structure by default, but also supports B+ Tree, Hash Table, and R-Tree for this purpose. It can also create secondary indexes. In addition, it can create an index in parallel to reduce the time needed. All indexes are stored on disk.

Storage Architecture

Disk-oriented

Although it is currently branded as the "Actian X Hybrid Database", the term "Hybrid" refers to its capability of performing both OLTP and OLAP tasks by employing a hybrid storage model (i.e. both row and column), not that it has a hybrid storage architecture. It is still disk-oriented.