Google Cloud 为 AI 优化运营数据库_AI阅读总结 — 包阅AI

As AI adoption continues to accelerate, operational databases provide the foundation for building enterprise AI apps that are accurate, relevant, and grounded in enterprise truth. Our goal at Google Cloud is to deliver the best databases for building and running AI apps. To that end, we’re excited to announce several new capabilities that make it easier for you to build intelligent gen AI apps with Spanner including Spanner Graph, vector search, and advanced full-text search. We’re also transforming the developer experience with the announcement of Bigtable SQL and Bigtable distributed counters, making it even easier to build at-scale applications. Finally, we’re making major announcements to support customers with their traditional enterprise workloads including SQL Server and Oracle, helping them modernize their data estates. Let’s dive in!

Easily build intelligent gen AI apps

Over the past year, we have been focused on helping developers build enterprise gen AI applications by providing industry-leading vector support and strong integration with Vertex AI and open-source LangChain. But we’ve also heard from customers that in order to build intelligent AI applications, they want to reason about knowledge — not just the data itself but how the data is interconnected. They also need to make it easy for their users to search for the right data — not only via keyword search, but also by implementing AI-enabled semantic search. That’s why we innovated on multiple new game-changing capabilities in Spanner, our always-on, globally consistent, and virtually unlimited scale database.

Today we’re announcing Spanner Graph which expands Spanner’s multi-model capabilities to include graph processing at virtually unlimited scale. Spanner Graph supports the industry-standard graph query language (GQL) and provides interoperability with SQL for seamless querying of structured and connected data in a single operation. Developers can now supercharge AI applications with knowledge graphs and Graph-based Retrieval Augmented Generation (GraphRAG), retailers can implement smarter recommendation engines, financial services firms can deliver sophisticated fraud detection, as well as many other high-value use-cases.

“Google Spanner’s multi-model and graph capabilities enhance our fraud detection and transactional data management, ensuring robust and reliable systems.” – Hai Sadon, Data Platform Group Manager, Transmit Security

“Spanner Graph’s flexible data modeling and high-performance querying have made it far easier for us to leverage the vast amount of data we have in our online applications.” – Aaron Tang, Senior Principal Engineer, U-NEXT

Keyword search and semantic search are also important building blocks for AI apps. Today, we’re announcing full-text search and vector search in Spanner. The new full-text search feature builds on Google’s decades of expertise in search and delivers highly scalable advanced full-text search.

Spanner’s new approximate nearest neighbor vector search is based on Google’s innovative ScaNN algorithm, which we first brought to AlloyDB and now Spanner. With it, you can now index and search vector embeddings to power AI-driven semantic search. With Spanner Graph, full-text search and vector search, we have evolved Spanner from not only being the most available, globally consistent and scalable database, to a multi-model database with intelligent capabilities that seamlessly interoperate to enable you to deliver a new class of AI-enabled applications.

Another recently launched capability — dual-region configurations — lets you take advantage of Spanner’s industry-leading 99.999% availability while complying with data residency requirements in Australia, Germany, India, and Japan. Spanner’s geo-partitioning enables you to retain the manageability of your single, global database while optimizing your costs and improving latency for your users who are distributed around the globe.

With all these new Spanner capabilities, we are also announcing Spanner editions, a new way of packaging and pricing Spanner that delivers more flexibility and cost transparency to fit every organization’s needs and budget.

Transform the developer experience

Bigtable is our wide-column key value store that is used by many Google services such as Google Ads, while customers such as PLAID and Mercari love Bigtable for its flexible schema, write throughput, and low latency. We’re committed to transforming the developer experience. In the past, developers had to use Bigtable APIs to take advantage of its data model and performance, even though SQL is still the go-to language for most data-driven applications. But that’s all about to change!

We’re thrilled to announce Bigtable SQL support. Developers can now leverage Bigtable’s performance, flexible data model and scale to build real-time applications using their existing skills. This is arguably the biggest change to Bigtable since it was first launched over 20 years ago. Starting today, more than 100 SQL functions are available directly in Bigtable. From kNN for developing gen AI applications and JSON manipulation for log processing, to using data sketches for real-time analytics, it’s easier than ever to build real-time, high-performance applications that can scale with your needs.

“Seamless SQL integration and efficient counter functionality in Bigtable will empower us to build more robust and scalable solutions for our customers.” – Jun Kusahana, VP of Engineering, PLAID

We also recently introduced Bigtable distributed counters to make it even easier to build at-scale applications with real-time embedded analytics, and today, we’re pleased to announce that distributed counters are now generally available. Distributed counters are optimized for high throughput, write-time aggregations such as processing event streams to support AI, fraud detection, data mesh, and recommendations use cases. Bigtable counters makes it easy to sum daily user engagement, find the minimum to determine the lowest sensor reading, use a maximum to identify your peak usage, or track an approximate count of distinct users with a backload of historical data from BigQuery.

Of course, we haven’t forgotten about data scientists. Building data pipelines and training large ML models on Bigtable will also get easier with new Spark connectors that support PySpark, Scala and SparkSQL, generally available now.

Modernize your database estate

Just last year, we announced the Enterprise Plus edition for Cloud SQL for PostgreSQL and MySQL and today, we’re excited to share that we are bringing Enterprise Plus edition to Cloud SQL for SQL Server. Cloud SQL Enterprise Plus edition for SQL Server delivers up to 4x improved read performance compared to the Cloud SQL Enterprise edition, offers higher availability with 99.99% SLA, and advanced disaster recovery.

“As a real-estate rental management service, we require better performance and availability for our database workloads, especially during peak business season. We will use the new performance-optimized machine in Cloud SQL for SQL Server Enterprise Plus edition to run our high-throughput workload and to benefit from the higher memory available per vCPU. Cloud SQL for SQL Server Enterprise Plus edition delivers the price-performance we need to run our rental management service that delights our customers.” – Shun Watanabe, CTO, Visual Research

Finally, to help accelerate your cloud transformation, we recently announced an exciting new strategic partnership with Oracle. We will be hosting Oracle database services like Oracle Exadata and Autonomous database services right within Google Cloud data centers. In addition, customers can also choose to run many Oracle database and application use-cases on Google Compute Engine or cross-connect between Oracle Cloud Infrastructure and Google Cloud. We are excited about how this partnership enables our joint customers to easily migrate their Oracle-based applications to Google Cloud.

Embrace an AI-driven future

There’s a wealth of data in operational databases just waiting to power the next transformative application. All these capabilities we’re announcing today will help you get the most out of your data.

To learn how to get started, visit our product page.