Author | Nuka-Cola, Li Dongmei
In the early morning of May 3, 2024, Beijing time, Oracle Corporation announced that it is providing the latest long-term support version of its database products, Oracle Database 23ai. It's worth mentioning that the database used to be called Database 23c, but it was later changed to its current name due to the addition of some AI features to the database.
In addition to AI capabilities, the cloud version of Oracle Database 23ai offers more than 300 new features, including Java Script storage procedures, priority transactions, data use case domains, schema permissions, Boolean data types, developer roles, JSON Schema, real-time SQL planning management, improved microservices support, XML and JSON search index enhancements, and more.
"For the past four years, Oracle Database Development has been working on the next long-term support release for Oracle Database, with a focus on AI and developer productivity," Oracle said in a press release. Given the strong focus on AI in the new version of the database, we have decided to change the name from Oracle Database 23c to Oracle Database 23ai to reflect the focus of this announcement and the overall context of the current technology market."
Oracle Database 23ai focuses on three key areas: Data & AI, Data & Development, and Data & Mission Critical.
Data & AI
In this section, Oracle Database 23ai's core goals are twofold: first, to make it easier for application developers to add AI capabilities to data-driven applications; The second is to incorporate generative AI capabilities into our products so that all users of Oracle Database, from data analysts to application developers and DBAs, can enjoy increased productivity.
The efficiency gains come from using AI to simplify data management for Oracle Database, providing AI algorithms to discover new insights from data, facilitating natural language interaction with database queries, and enabling users to store AI-generated vectors for quick similarity searches across objects.
Enhance next-generation AI models
Over the past two years, the impact of AI's large language models (LLMs) has been felt, as exemplified by products such as ChatGPT, Google Gemini, Cohere, and Lamama. While large models have a profound impact on the way people do their day-to-day work, these engines also face the same challenge – they can only answer questions within the specific time window in which they were created. That is, they don't make effective use of the data, context, and more details that are held within the organization. In order to make various large models play a more comprehensive role, it is necessary to combine the existing data in the database to help and enhance the ability of large models to answer relevant questions.
In Oracle Database 23ai, the Oracle R&D team introduced AI vector search. It's a powerful new technology that allows you to generate and store vectors with next-generation AI models. These vectors (also known as embeddings) are multi-dimensional representations of documents, images, videos, sounds, and more. By encoding these objects as vectors, mathematical calculations can be used to find similarities between them.
The real power of Oracle Database 23ai is that it combines these similarity searches with business data searches using simple SQL. That said, anyone with a basic understanding of SQL can write powerful statements that combine similarity with other search criteria. These kinds of queries bring additional context to the larger model, enhance their knowledge, make their answers more accurate, and relevant to the content of the customer's or organization's question. To enable this feature, developers have also added new data types, new vector indexes, and SQL language extensions to lower the barrier to entry for advanced analytics, query vectors, and existing business data in Oracle Database 23ai.
So, how do these features come about?
According to Oracle, in order to realize the potential benefits of large language models, they first need to deepen their understanding of the dataset and the objects within it. The models that encode these objects are often built from a large and complex myriad of code fragments. As a result, organizations often need to hand over processing to a third-party service. This sharing of potentially sensitive information with third parties naturally brings risks.
Data Security Coding
To ensure secure encoding of objects, Oracle Database 23ai allows users to load their own trusted AI models directly into the database with ONNX standard support. This allows Oracle Database 23ai to encode objects as they are inserted into the database. This not only enables near-real-time secure inference on objects, but also avoids the risk of extracting data from a database and handing it off to a third-party service provider.
Ask questions in a natural and fluid way
While SQL is a very powerful language, it's also quite difficult to get started. To enable more users to ask complex questions about Oracle Database, we're working on integration with large language models such as Cohere and Llama to enable natural language support for the questioning process. Now users can simply ask questions such as "Shows the sales of the most popular products among younger consumers in the last 4 quarters." Oracle Database 23ai will share in-table metadata with large models to answer user questions. The large model understands what "younger generation" means and converts it into a range of data that can be used as a filter in a SQL query to return instructions to the database. Of course, in addition to querying the information in the database, Oracle Database 23ai also allows users to ask questions about various aspects of the content.
Since the release of Oracle Database 9iR2 more than 20 years ago, the database has been built with machine learning (ML) algorithms to help users quickly find patterns, trends, and predict customer behavior in the data within tables. These machine learning algorithms allow you to easily perform a variety of functions without having to go through complex ETL operations to extract data and write back the results. These machine learning models can also be used for classification, clustering, and prediction when data is inserted or loaded into an Oracle database, providing a range of tangible business benefits, including customer recommendations and fraud detection. Since this release, the R&D team has improved Oracle Database and added new machine learning algorithms and features to build one of the most sophisticated and powerful data mining platforms in the industry.
Available for all Oracle Database editions
AI Vector Search will be available at no additional charge in Enterprise Edition, Standard Edition 2, Database Free, and all Oracle Database Cloud Services as part of Oracle Database 23ai.
AI Vector Search has garnered a lot of attention before its release, and many customers and partners are keen to try it out and provide feedback on the latest features such as AI Vector Search. Heli Helskyaho, CEO of Miracle Finland Oy, said: "AI vector search has changed the way we look at our business by bringing AI into data: for example, we can use natural language to query document content to find answers to customer questions. This makes the process of serving customers faster and the quality of service improved. This combination of the power of generative AI with your own data is set to make a huge difference. ”
Data & Development
Oracle Database 23ai simplifies the application development experience by eliminating the complexity associated with database interactions. Removing complexity from the application development process means that users can focus more on building more elegant applications and less on the technical details. In addition, reducing complexity can help shorten development cycles and help users better adapt to today's digital environment, where market demands are changing rapidly and development rhythms are difficult to control.
JSON or relationships – a tough choice
Oracle 23ai introduces a number of key technologies to reduce the complexity faced by developers, but one of the most innovative and effective is the introduction of JSON-relational duality. This technology allows users to take advantage of the power of relationships and the convenience of a JSON development approach in a single application. JSON provides an elegant approach to data modeling, where all the relevant data needed to answer a query is contained in a single object, without the need to perform complex joins across tables. The relational approach provides a flexible, store-efficient, and consistent data model that is easy to query in languages such as SQL. Historically, users have been forced to choose a database modeling approach at the beginning of the application lifecycle, and decisions made at an early stage have the potential to have far-reaching or even hinder business success as business needs change. The advantage of JSON-relational duality is that users are able to enjoy the benefits of both approaches without having to use a complex and clunky object-relational mapping framework (ORM).
By creating a simple JSON Duality view on top of a relational table, users can build applications that use documents to query and update the underlying dataset. Not only does this avoid data duplication, a fundamental problem in document modeling, but it even hides the complexity of database-level concurrency control, with Oracle Database 23ai managing document-level seriality on behalf of the user. You can continue to use simple HTTP PUT, POST, and GET operations, or use language-specific APIs such as the Oracle SODA API or even the MongoDB API directly against Oracle Database. With JSON Duality Views, "you can't have it both ways" is a thing of the past.
Make SQL easier to use
SQL remains one of the most popular development languages among today's application developer community. This ability to query business data, from the simplest to the most complex, makes SQL an absolute cornerstone of data management. For the foreseeable future, SQL's status as a charlatan is likely to remain unshakable. Of course, there are still areas for improvement. In Oracle Database 23ai, developers listened to a large number of enhancement requests and looked at what could actually simplify or make database interactions more compelling. Improvements to SQL include the addition of two new data types, Boolean and Vector, the addition of Data Use Case Fields to allow users to enrich the meaning of the data types used in table definitions, the removal of the hard requirement of "from Dual" for simple SQL statements to improve readability, making it easier to aggregate "interval" data types, and the introduction of table-valued constructors to easily specify multiple rows in insert select or merge statements.
Model attribute graphs for complex relationships
Graph databases provide a new perspective on modeling complex relationships in real-world scenarios. Until now, however, developers have struggled to solve the problems posed by proprietary languages and have had to send data to a database dedicated to managing these relationships, thus hindering the widespread adoption of this breakthrough technology. Fortunately, the database industry is trying to democratize graph database technology in a collaborative way, and eventually establishing a standard. These standards enable developers to familiarize themselves with and take advantage of the breadth of SQL's capabilities, resulting in a more accessible environment. Oracle Database 23ai is the first commercial database built using this new standard. With these advancements, it is expected that the integration of graph databases in the enterprise ecosystem will see significant growth.
As with the JSON binary view, the property graph view allows you to express how you want to use an existing relational table, or how to use data that is already saved in a JSON cluster. With a simple view, users can enrich their data to model the potentially complex relationships contained within the data. The new SQL extension also simplifies the process of writing powerful queries, allowing users to gain deeper understanding of information that was previously difficult to describe with standard SQL. The developers have also added a new set of complex graph algorithms (shortest path, classification, joining, etc.) that will be available as soon as the GA release is released.
Property Graph 属性图将在 Oracle Database 23ai 的所有版本中向用户开放。
Data & Mission Critical
Oracle Database has a long history of supporting the world's most critical use cases, with an unmatched breadth of capabilities and uses, and an expertise at protecting critical systems from unplanned downtime. A typical use case is Oracle Real Applications Clusters (RAC), a cutting-edge technology that enables live, shared-disk databases for mission-critical scenarios. Oracle Active DataGuard and Oracle GoldenGate also offer unique flexibility in building homogeneous and heterogeneous data replication architectures.
Make sharding better, faster, and simpler
For use cases where enterprises want to distribute data over longer distances, whether due to hard regulatory requirements that user data must be stored in the customer's country or to reduce the latency associated with globally distributed applications, the developers also introduced database sharding from Oracle Database 12. Sharding is the distribution of a dataset across multiple Oracle databases, but managed as a single logical data set. This sharding architecture allows the database to be scaled up or down by adding and removing shards, and allows Oracle to automatically rebalance the dataset. Oracle is able to do this online, and because the sharding mechanism is embedded directly within the database, there is little impact on the uptime of the business.
In Oracle Database 23ai, globally distributed database technology introduces support for Raft replication. It's a consensus-based protocol that helps automate the configuration of replication across all shards. With Raft, replicated data is saved as a replica of a "replication unit" across all shards. In the event of a shard failure or reconfiguration, Oracle Database 23ai automatically rebalances the data. The new Raft protocol also supports sub-second failover for business continuity.
Oracle 23ai also brings a number of comprehensive improvements to the high availability product, such as a new "Local Rolling Database Maintenance" that simplifies the impact of rolling patches in a cluster by establishing another instance on the same node that receives the patch. Oracle has also introduced new Data Guard Per Pluggable Database (PDB) support, which allows pluggable database-level disaster protection between two container databases (CDBs), each running an active workload. We've also improved PDB integration and management in Oracle Clusterware.
Caching data made easy
Caching data at the application layer has a number of benefits, including improving application responsiveness and reducing database load as queries are moved to a cache closer to the code. However, in order to realize these benefits, developers often face a number of challenges, one of the biggest of which is ensuring that the cache is consistent with the data held in the back-end database, which risks providing stale/potentially incorrect data. Historically, developers have been forced to write complex code to manage these scenarios, often resorting to design patterns such as "time-to-live," where data is periodically refreshed and reloaded within the cache. The complexity of these challenges means that caching can be difficult and expensive to maintain. In some cases, maintenance costs are often higher and more user-friendly than design.
In Oracle Database 23c, developers want to solve the caching problem in Oracle Database with a new feature called "True Cache". True Cache is a non-disk Oracle instance that runs before the primary database and leverages Active Data Guard technology to ensure that it is automatically updated as data changes in the primary database instance. One of the most important features of this new caching technology is that it works with minimal configuration and minimal code changes.
Make SQL execution more secure
One of the biggest challenges organizations face is that malicious actors are constantly trying to steal data and disrupt their day-to-day operations. In Oracle Database 23ai, the developers introduced the SQL Firewall to prevent the execution of unauthorized SQL, including SQL injection attacks and actions by unauthorized persons. SQL Firewall allows users to train a database with SQL that is allowed to be executed. The SQL's Approved List then records the unauthorized SQL or its execution. Users can also add additional conditions to explicitly list exceptions in IP ranges or authorized callers. SQL Firewall is so powerful because it's part of Oracle Database, so the additional overhead for SQL execution is very limited. SQL Firewall is very simple to set up and train, and can be done easily through the GUI or with a simple call to a stored procedure.
In fact, the main reason why the release of Oracle Database 23ai has attracted so much attention is because of the introduction of Oracle AI vector retrieval. This feature allows customers to easily search documents, images, and other unstructured data based on conceptual content rather than specific words, pixels, or data values.
Oracle Database 23ai is especially important in the area of data security by securely combining searches for unstructured and private business data without moving and replicating data.
Unlike traditional AI algorithms, which need to move data where the algorithm resides, Oracle Database 23ai brings AI algorithms to where the data is. This enables AI to run in real-time in Oracle Database, improving the effectiveness, efficiency, and security of AI applications.
Original link: More than 300 new features, Oracle Database 23ai is officially released! The database giant, which is nearly 50 years old, is entering the AI era_Data lakehouse_Nuka-Cola_InfoQ featured article