Chapter 01 · Oracle AI Database Foundations, Editions, Deployment Models, and Lab Setup

Oracle Database Architecture, Converged Database Model, Workloads, and Product Terminology

Build a precise Oracle AI Database 26ai mental model: database/instance and CDB/PDB boundaries, listener/services, converged capabilities, and observable architecture evidence.

Intermediate95–115 minutesArchitecture + evidence labOracle AI Database 26ai baselineFree + SQLcl/SQL*PlusLast reviewed: August 24, 2026

Learning outcomes

The fictional ServiceHub field-service platform already knows relational SQL from earlier academy courses. Its next deployment must support transactional work orders, JSON payloads from mobile clients, location-aware dispatching, and eventually vector-assisted knowledge search. The team has chosen Oracle for a proof of concept, but the phrase “Oracle database” is being used for everything from a listener process to Exadata. That ambiguity is dangerous: you cannot troubleshoot, secure, recover, or license a system whose boundaries you cannot name.

01

Explain the difference between Oracle database files, an Oracle instance, a container database (CDB), and a pluggable database (PDB).

02

Trace a client connection through a listener and service to a database session without confusing a service name with an instance identifier.

03

Place SGA, PGA, redo, undo, background processes, schemas, and tablespaces in a first-pass architecture map.

04

Explain Oracle’s converged database model while keeping Autonomous AI Database, Exadata, RAC, and GoldenGate as distinct products or architectures.

05

Collect read-only architecture evidence before making an operational claim.

Prerequisite connection

Courses 01 and 02 established relational modeling and portable SQL. SQLite showed an embedded database, while MySQL, PostgreSQL, MariaDB, and SQL Server introduced server processes and engine-specific administration. Carry those mental models forward only as comparison points: Oracle’s instance/database and CDB/PDB boundaries are its own.

1. Database and instance are different things

In Oracle terminology, a database is fundamentally the persistent set of files and metadata that hold durable state. An instance is the running memory structures and operating-system processes that access an Oracle database. On a conventional single-instance deployment, one instance opens one database. Oracle Real Application Clusters (RAC), introduced much later in this course, deliberately changes that topology by allowing multiple instances to open the same database. That is why “instance” and “database” are not synonyms even when a small lab makes them look one-to-one.

The central shared-memory region is the System Global Area (SGA). Each server process also uses private process memory called a Program Global Area (PGA). Background processes coordinate writing dirty buffers, redo, checkpoints, cleanup, and other work. You do not need to memorize every process now. The useful first model is: persistent database files survive process restarts; the instance is the running machinery that opens and operates those files.

sql · identify the running instance and database
SELECT banner_full FROM v$version;SELECT instance_name,       host_name,       version,       version_full,       status,       database_statusFROM   v$instance;SELECT name,       dbid,       open_mode,       database_role,       cdbFROM   v$database;

Run these queries as a local administrative lab account with access to the dynamic performance views. The values describe different layers. V$INSTANCE describes the running instance; V$DATABASE describes the database opened by that instance. A current 26ai Free package may report a VERSION_FULL in the 23.26.x train. Record the exact value you observe instead of replacing it with the marketing name “26ai.”

2. CDBs and PDBs add another boundary

Modern Oracle deployments use the multitenant architecture. A container database (CDB) contains a root container, CDB$ROOT, a read-only seed used as a template, and one or more pluggable databases (PDBs). A PDB is the application-facing database boundary in which ordinary local users and application schemas normally live. Oracle AI Database Free creates a CDB named FREE and a default PDB named FREEPDB1.

A schema is the namespace of objects owned by an Oracle database user. It is not another database process. A tablespace is a logical storage container backed by datafiles. These terms answer different questions: “which container is my session using?”, “who owns this table?”, and “where is this segment allocated?” should never be collapsed into one concept.

sqlcl / sql*plus + sql · prove the current container and PDB state
SHOW CON_NAMESHOW USERSELECT sys_context('USERENV','DB_NAME')       AS db_name,       sys_context('USERENV','CON_NAME')      AS con_name,       sys_context('USERENV','SERVICE_NAME')  AS service_name,       sys_context('USERENV','CURRENT_SCHEMA') AS current_schemaFROM dual;SELECT con_id, name, open_mode, restrictedFROM   v$pdbsORDER  BY con_id;

SHOW CON_NAME and SHOW USER are client commands understood by SQL*Plus and SQLcl; the SELECT statements are server-side SQL. That distinction matters throughout the course. A session connected through FREEPDB1 should normally report that PDB as its current container and service, while a privileged connection to the root service reports CDB$ROOT.

3. Listener, service, session, and SID solve different problems

A remote client normally contacts an Oracle Net listener. The listener receives the initial network connection and uses registered services to route the client toward the appropriate database workload. A service is a logical connection target and workload-management name. It is deliberately more portable than hard-coding a process identity. After handoff, the database session does not send every SQL statement “through the listener” as if the listener were a query proxy.

SID historically identifies an instance on a host and still matters to local administration and some connection configurations. It is not the same as a service name. For the default Free installation, FREE is the CDB/instance naming baseline while FREEPDB1 is the application-oriented PDB service. Later chapters will make service registration, dedicated/shared server, RAC services, and Data Guard service behavior precise.

shell · observe listener registration outside SQL
# Run on the database host or inside the database container.lsnrctl status# Look for listening endpoints (commonly TCP/1521 in the default Free lab)# and registered services such as FREE and FREEPDB1.

4. SGA, PGA, redo, undo, and storage: a first-pass map

The SGA contains shared structures used by many sessions, including the database buffer cache and shared SQL areas. PGA is private to a server or background process and is used for session/process state and work areas. Redo records changes needed to reproduce database modifications during recovery; undo records information needed to reverse transactions and reconstruct consistent older versions for readers. They solve different problems. Neither is “a second copy of the table.”

Physical persistence includes datafiles, control files, and online redo logs. Logical storage introduces tablespaces, segments, extents, and blocks. Chapter 02 will open the instance-memory/process model, Chapter 03 will open storage, and Chapters 07 and 14 will explain transaction consistency and redo/undo deeply. In this lesson the goal is orientation: know where a term belongs before tuning it.

sql · collect read-only memory and file evidence
SELECT name,       ROUND(value/1024/1024, 1) AS mbFROM   v$sgaORDER  BY name;SELECT file#, nameFROM   v$datafileORDER  BY file#;SELECT group#, thread#, sequence#, bytes/1024/1024 AS mb, statusFROM   v$logORDER  BY group#;

These queries do not prove that the current memory sizes or redo-log sizes are “good.” They only establish observable state. Tuning requires workload evidence, platform limits, and change validation; this course will not turn catalog values into folklore recommendations.

5. Oracle’s converged model—and the product names it does not erase

Oracle describes the database as converged because multiple data models and workload capabilities can live beside relational tables and participate in SQL. The current product family includes native JSON, spatial capabilities, graph-related functionality, and vector/AI features in addition to relational SQL and PL/SQL. A ServiceHub work order can therefore keep strongly typed relational columns while storing a JSON device payload, spatial coordinates, or a vector embedding when the data model justifies it.

Name What it is Do not confuse it with
Oracle AI Database The core database product and engine family used by this course. A specific cloud service or hardware appliance.
Autonomous AI Database An Oracle-managed cloud database service with automated operational responsibilities. A synonym for every Oracle database.
Exadata An engineered database platform and cloud/on-premises system family. The Oracle SQL language or ordinary database instance.
Oracle RAC A clustered architecture in which multiple instances can access one database. Data Guard disaster-recovery replication.
Oracle GoldenGate A separate data replication/integration technology. Redo-based crash recovery inside one database.
Oracle AI Vector Search Database vector data/search capabilities. An external large-language model or an automatic replacement for relational design.

The converged claim is an architectural choice, not a requirement to store every data type in one product. Later lessons will compare native capabilities with specialized external systems using workload evidence, operating cost, governance, and failure domains.

6. Deliberately wrong model: “FREEPDB1 is the instance”

A new operator sees FREEPDB1 in the connection string and writes: “The instance is FREEPDB1 and the database is the listener on 1521.” Nothing in that sentence has a stable boundary. The port belongs to a listener endpoint, the service routes a connection, the PDB is a container inside the CDB, and the instance is the running SGA/process set.

sql · repair the model with one evidence card
SELECT i.instance_name,       i.host_name,       i.version_full,       d.name AS database_name,       d.cdb,       sys_context('USERENV','CON_NAME') AS current_container,       sys_context('USERENV','SERVICE_NAME') AS service_name,       sys_context('USERENV','CURRENT_SCHEMA') AS current_schemaFROM   v$instance iCROSS JOIN v$database d;

Interpret each column independently. If you later clone or relocate a PDB, the service/container can change while the host and instance are different facts. If RAC is introduced, multiple instances can open the same database. The repaired incident note therefore records instance, database, current container, service, host, and exact RU separately.

7. Hands-on lab: build the Chapter 01 architecture evidence card

Use Oracle AI Database Free 26ai locally, or Oracle Live SQL for SQL-only exploration when host/listener evidence is not required. For local administration, connect only to a disposable learning environment. Do not change initialization parameters or storage yet.

sql · architecture evidence card
SELECT systimestamp AS observed_at FROM dual;SELECT banner_full FROM v$version;SELECT instance_name, host_name, version_full, status, database_statusFROM   v$instance;SELECT name AS database_name, dbid, open_mode, database_role, cdbFROM   v$database;SELECT sys_context('USERENV','CON_NAME') AS con_name,       sys_context('USERENV','SERVICE_NAME') AS service_name,       sys_context('USERENV','SESSION_USER') AS session_user,       sys_context('USERENV','CURRENT_SCHEMA') AS current_schemaFROM dual;SELECT con_id, name, open_modeFROM   v$pdbsORDER  BY con_id;SELECT value AS compatibleFROM   v$parameterWHERE  name = 'compatible';

Verification checklist:

  • You can state the exact observed VERSION_FULL and the configured COMPATIBLE value without treating them as the same thing.
  • You can identify the instance name, CDB name, current PDB/container, service name, session user, and current schema.
  • You can describe the listener as a connection-establishment component rather than the database engine.
  • You can place SGA/PGA, redo/undo, tablespaces/datafiles, and schemas in the correct layer.
  • You can name which parts of your lab are local Oracle Database behavior and which would require cloud, Exadata, RAC, or GoldenGate infrastructure.

8. Production judgment

Good Oracle diagrams show ownership and failure boundaries. Record the database software/RU, Oracle Home, instance, CDB/PDBs, services/listeners, storage, credentials, backups, and any cluster/cloud control plane separately. When a feature belongs to a licensed option, management pack, engineered system, or cloud service, label that boundary instead of assuming it is part of “Oracle.”

From this point forward, every operational statement should answer: what scope is this? what exact RU and COMPATIBLE level apply? which edition/offering/option entitlement applies? what privilege and container am I in? and what evidence proves the claim? Lesson 2 turns those questions into a repeatable version and licensing discipline.

9. Summary and next step

An Oracle database is persistent state; an instance is the memory/process machinery that operates it. Multitenant adds CDB/PDB boundaries, services route clients to workloads, schemas own objects, tablespaces organize storage, and the SGA/PGA plus redo/undo participate in runtime and recovery. Oracle’s converged features live inside this product family, while Autonomous, Exadata, RAC, and GoldenGate remain distinct architectures or products. Next, you will separate the 26ai product name from its 23.26.x RU version train and learn why feature presence is not proof of production entitlement.

Check your understanding

  1. Why is “database = instance” an unsafe mental model even in a single-instance lab?
  2. What is the practical difference between a PDB service name and an SID/instance identifier?
  3. Why does querying V$SGA tell you current state but not whether the memory configuration is optimal?
  4. Name two Oracle products or architectures that must not be treated as synonyms for the core database engine.
  5. What five kinds of evidence should appear in an Oracle architecture incident note before diagnosis begins?
Review the answers

The database is persistent files/state, while the instance is running memory and processes. RAC makes the distinction explicit because multiple instances can open one database.

A service is a logical workload/connection target that can route clients; an SID identifies an instance context on a host. In multitenant deployments applications should normally target a PDB service.

V$SGA reports allocated shared-memory components. It contains no workload objective, latency baseline, concurrency requirement, or proof that a different allocation would perform better.

Examples include Autonomous AI Database, Exadata, Oracle RAC, and GoldenGate. Each has its own operational or topology boundary.

Record the exact RU/version, instance/host, database/CDB, current PDB/container, and service/session context; then add storage/listener/topology evidence relevant to the incident.

Authoritative references

Keep knowledge open

Help the academy stay free and grow.

If these tutorials save you time, a small donation supports new lessons, technical review, diagrams, examples, and long-term maintenance.

ETHEthereum / ERC-20 only
0x716c4Ab160C4B66F31a28AE2448BfF68fc3a2ef0

Send only Ethereum or ERC-20 compatible assets to this address.