Google wants to takes a byte out of Oracle workloads with PostgreSQL migration service

Google is looking to capture some of Oracle's workloads by promising to help users migrate from Big Red's databases to PostgreSQL with a set of services and automation tools.

Specifically, the Chocolate Factory wants to take users from Oracle databases to its managed PostgreSQL-compatible database service, AlloyDB. Among a slew of announcements at Google's Cloud Next event in San Francisco – which included generative AI, TPUs, GPUs, and Kubernetes – was the promise it could speed up Oracle migrations.

On a call last week, Andi Gutmans, veep and general manager for databases at Google Cloud, said: "Customers are looking to modernize off their legacy databases like Oracle and SQL Server, and they want to embrace more modern open technologies as part of their innovation roadmaps."

Google has launched the general availability of its database migration service for Oracle to PostgreSQL, which promise a "much easier migration path for customers who want to modernize," Gutmans said.

Google also announced the preview of Duet AI in Database Migration Service, which proposes to co-opt its generative AI chatbot to "help with that last mile of code conversion, where all the traditional methods failed," he said.

Gutmans said Chicago Mercantile Exchange, the global derivatives marketplace, is looking to AlloyDB for its most demanding enterprise workloads and is in the process of migrating several databases from Oracle.

According to this blog post, the preview of the AI-assisted migration will "automate the conversion of Oracle database code such as stored procedures, functions, triggers, packages and custom PL/SQL code, that could not be converted with traditional translation technologies."

Prompted by The Register to elaborate, a Google spokesperson said one of the biggest challenges in migrating from Oracle to PostgreSQL/AlloyDB is reproducing the underlying PL/SQL code, the procedural language Oracle uses to embrace SQL statements within its syntax.

"Oracle PL/SQL is really sophisticated and customers have built large complex applications based on this dialect for a long time. First of all, translating this PL/SQL to ensure syntactic correctness is relatively easier as compared to semantic conversion. In Database Migration Service, schema and code conversion are integrated so users have one place to go.

"With Duet AI assist, there are pre-trained models that are used for code conversion. We do not train our models with customer data. Duet AI helps the last mile conversion based on few manual edits. These edits are recorded and used to train the models to help with recommendations and automation.

"For example, SYSDATE from Oracle does not have an equivalent in PostgreSQL. With Duet AI powered code conversion, when the user manually adjusts the SQL, the recommendation is recorded and can be replayed and applied to other scenarios. All the user actions around code edits are recorded and used to refine the LLM, which is then used to provide future recommendation and search/replace capabilities. Users can move any PL/SQL applications that rely heavily on stored procedures, functions and triggers."

• Google throws down gauntlet with first compute instances powered by AmpereOne chips
• Google threatens to inject Duet AI bot into more corners of Workspace: Meet, Chat, etc
• Google sharpens AI toolset with new chips, GPUs, more at Cloud Next
• Nvidia just made a killing on AI – where is everyone else?

Google has also produced AlloyDB AI, a PostgreSQL-compatible database service to help combine LLMs with their real-time operational data by supporting vector embeddings. Other database providers including MongoDB, DataStax's Cassandra database service Astra, and open source PostgreSQL (via Pgvector) already support vector embeddings, while others advocate using a specialist vector database such as Pinecone.

Meanwhile, Google has moved AlloyDB Omni to public preview, offering the database on-prem and in rival cloud infrastructure including AWS and Azure, or "even on developer laptops."

Doug Henschen, vice president and principal analyst with Constellation Research, said AlloyDB should be taken as whole as it was "crucial for Google Cloud to gain traction with this database service, which is Google Cloud's alternative to Oracle and something they need to gain the same sort of customer traction that AWS has seen with Aurora."

However, he was more skeptical about the migration service. "Any and all database migration capabilities and services are certainly helpful and welcome, but Oracle Database tends to be pretty sticky and certainly tightly woven into any and all Oracle application deployments. Google Cloud really has to stick to this work and hope for customers to move third-party (non-Oracle) app deployments, new app deployments, and next-generation custom applications onto Google Cloud with AlloyDB. Heavily customized legacy enterprise app deployments and, particularly, Oracle apps, are likely to keep running on Oracle Database."

Google is not alone in trying to snipe Oracle workloads. In May, EDB launched a so-called "risk free" Oracle migration service in which customers signing a two-year contract to run a 64-core instance of PostgreSQL would not start paying until the new system is up and running and has been tested. ®