Guest post from Carl Olofson, Principal Analyst, DBMSGuru LLC

Overview

Deploying enterprise-level AI functionality can represent quite a challenge. We all like to think that our businesses and their immediate environment float on a sea of immutable and unambiguous facts. We should be able to ask any question that is rational and clear, and get an answer equally rational, clear, and accurate. So, layering AI over the top should be a simple proposition: install the software, configure, and game over. There are three problems with that expectation: one is that the information that surrounds us is not unambiguous, two is that our questions often are ambiguous, and three is that even if the first two were not an issue, we don’t have immediate access to the data necessary to answer many of the questions we would like to ask.

We can’t do anything about the information arising from data outside of our control, though with AI we can find probable answers to our questions, assuming we can access the data. The data that is within our control ought to be manageable enough, but often it is not. It resides in myriads of unconnected systems, and is not only uncoordinated, but is often not well enough defined or documented to know where it is or what it means. As a result, pulling the data together that we need to make timely short-term decisions or meaningful long-range decisions becomes something of a treasure hunt, chewing up time and resources.

Getting the Most Out of AI

Enterprises have more databases and more data than ever before, and when it comes time to harvest intelligence from them, there are logistical and efficiency challenges. Establishing a uniform foundation, based on open standards, with integrated AI built in, can overcome those challenges. Suppose we could have a database platform that enables us to find the data we need when we need it, with high confidence in its consistency and timeliness? This is the essence of an enterprise AI database.

Such a database platform requires a database management system (DBMS) flexible enough to support multiple structural models, scalable enough so that we don’t have to keep reconfiguring when we run out of space, and smart enough to find what we need based on AI capabilities optimized by the DBMS itself. Having one core system enables optimization of data management and access across all the models and workloads being served. Having one core database platform does not suggest that it supports just one data model or framework, by the way. The user community is not well served unless all data models of interest are supported.

Oracle AI Database 26ai

At this year’s Oracle AI World in Las Vegas, Oracle presented its newest flagship product instantiation: Oracle AI Database 26ai. First, as a continuation of the Oracle Database product, it contains support for everything previously supported, including all the same mission-critical capabilities. It comes in a do-it-yourself management configuration or as the self-running, self-tuning Autonomous AI Database. It can be run on-premises, in the cloud in hyperscaler environments, and on Oracle Cloud Infrastructure (OCI). It supports all the most common data management models, including (of course) relational tables as well as JSON documents, graphs, and more. It also supports spatial data, immutable (blockchain) ledgers, and storing and analyzing data in text files. It can optimize data storage and access for a range of workloads including AI, OLTP, analytical, streaming, and IoT. It can maintain a database in one physical or cloud location or, with the distributed database feature, across many regions. And it can be run on bare metal, in containers, on-premises Oracle engineered systems or OCI using Cloud@Customer, and OCI in the Oracle Cloud, or on Azure, AWS, or Google Cloud.

But you knew all that. What about the new stuff? What makes Oracle AI Database 26ai an ideal platform for comprehensive data-driven AI deployment across the enterprise?

New Features

The following new features add to the list of capabilities that Oracle AI Database 26ai already has to serve as a premier enterprise AI data platform:

  • AI Vector Search, including with LangChain
  • Data Use Case Domains
  • Exadata AI Smart Scan
  • LLM Integration
  • Annotations
  • External Table Support for Vectors
  • Enhanced JSON Relational Duality Views
  • True Cache
  • Property Graph Support
  • SQL Firewall
  • Globally Distributed Database

In addition, the new release also features real-time SQL plan management, JavaScript stored procedures, read-only PDB standby, and rolling patching.

How Oracle AI Database 26ai Stacks Up as the AI Platform Core

Let’s look at how this DBMS, with its latest updates, addresses the requirements for an enterprise AI data platform:

  • Data can be searched on a similarity basis, or based on approximate characteristics using Oracle AI Vector Search, which can be incorporated into standard SQL and also used to search unstructured data. Unlike other solutions that perform vector search as a process outside the database core, Oracle integrates this functionality in the database system for maximum efficiency and flexibility. Finding and combining data is eased using Oracle Autonomous AI Database’s Catalog of Catalogs.
  • For databases physically distributed across regions, Oracle Globally Distributed Database provides both consistency and optimal transaction throughput and replication using the Raft algorithm.
  • Oracle SQL can be used to access data not just in Oracle Database but in others, with special emphasis on Apache Iceberg tables. Oracle True Cache overcomes performance differences with a dependable point-in-time caching function that is much more efficient than some external third-party solution.
  • With the JSON Relational Duality Views feature, access data stored in relational format as JSON document views, ensuring data consistency while simplifying application development. Also, Select AI generates the right SQL for any desired query based on natural language prompts.
  • Can scale dynamically as the amount of relevant data increases. For users of OCI, the database can scale up or down dynamically and transparently.
  • Enables the building of applications by non-technical users with natural language and point-and-click interfaces. Data can be handled as relational, JSON, or graph within the Oracle AI Database 26ai with no special expertise required.
  • In addition to AI features that greatly accelerate the development process for professional developers, this release also enables JavaScript developers to leverage their skills through support for JavaScript stored procedures.
  • Because the platform handles data in a comprehensive way it must have security features that are integrated into SQL processing. In-database Oracle SQL Firewall can block unauthorized SQL and SQL injection attacks and provides full visibility into all SQL traffic for review and fault detection.

Making AI Deployment Simpler, not More Complex: Put AI Inside, not Alongside, the Database

Many AI solutions involve using a LLM to process a prompt forming SQL that can run on a database that is separate from, or runs alongside, the AI system. This introduces both coordination and performance problems. Having the AI system running inside the database environment, as Oracle AI Database 26ai does, overcomes these challenges. It is useful to note that Oracle is applying its vision, AI for Data, to support all the leading AI models and frameworks, and the API technology can be called via APIs or deployed as private instances.

Iceberg Support

A popular option for sharing data across disparate systems is to build Apache Iceberg tables, but these usually suffer performance problems because they run unoptimized, on block or object storage. Oracle AI Database provides an optimized environment for Iceberg tables, so they show similar performance characteristics to Oracle tables, and their operations can be blended together. This enables users who create Iceberg tables to incorporate Iceberg tables into work done on the Oracle AI Database platform either alongside Oracle Database tables or separately.

Conclusions

Most enterprises have many application databases, and they are mostly unrelated. Without some means of managing all of the data with speed and scale, and without key AI features built-in, the dream of offering AI-driven access to any and all relevant business data remains just that: a dream. An enterprise AI database provides a way of rationally managing this data, managing large amounts of transactional data all together, exposing it for analytics, and supporting both application development and analytics using advanced AI technologies.

Oracle AI Database 26ai addresses these issues. It has the technology needed to manage large databases as well as large numbers of smaller database instances while classifying the data they hold and delivering it in combinations determined by both query and intelligent search, including similarity search, to large numbers of concurrent users using natural language. It can do these things on-premises or on each of the major hyperscaler platforms (including AWS, Azure, Google Cloud) in addition to Oracle Cloud. Anyone managing a data environment of any significance that is seeking to build AI applications needs a comprehensive database system to corral and bring that data to AI. Oracle offers Oracle AI Database 26ai as the ideal core technology for such an AI data environment.