Networking in the Google Cloud is continuing to move forward, with a series of innovations outlined recently at Google Next.

Google execs specifically outlined the hyperscaler's latest efforts to advance cloud networking. As was the case with the majority of Google Next this year, artificial intelligence (AI) was a primary theme, but it wasn't the only one. After all, while AI in the cloud is a growing trend, there are still vastly more non-AI workloads running in the cloud that can still benefit from optimized networking.

Back in September, Google introduced its Cross-Cloud Network multicloud capabilities at the Google Cloud Next 23 event.  That effort is now being expanded with the  Service-centric Cross-Cloud Network set of features.

“The cross cloud network is a new era of cloud networking,” Muninder Sambi, VP/GM of cloud networking at Google, said during a session at Next ’24.

The service-centric cross-cloud is a new approach introduced by Google Cloud to simplify connecting to services across different environments.

“We are really pleased with what Google Cloud can offer to help simplify and make it easy for you to deploy any sort of cloud to any service in a simple, reliable and secure manner,” Sambi said. “With this, we're introducing service centric cross cloud network that allows you to now connect to any workload that you have, whether it's for AI/ML workloads, Vertex, or any best of breed GCP services, from on-prem assets, or from any other cloud in a simple, easy, secure manner, and fully reliable, fully managed by Google Cloud.”

Sambi said that the service-centric cross cloud provides a consistent way for organizations to connect to services wherever they might be. He explained that it is like having a Private Service Connect (PSC) Virtual private cloud (VCP) tunnel that is optimized and makes it simple for DevOps to publish using traffic management for all types of workload rollouts.

The service integrates multicloud traffic management best practices including cross region failover, health checks, dynamic route propagation and security.

Google's Gemini generative AI large language model (LLM) is being put to task in service of networking.

Gemini Cloud Assist is a new tool that leverages the LLM and domain knowledge to improve various aspects of networking.

“Gemini Cloud Assist is much more than a raw language model, it will revolutionize the entire cloud lifecycle from designing your network to operating and troubleshooting your network to securing and optimizing your network,” Mark Church, product manager at Google said during the Next '24 session. “This is because it has access to your context and can reason about your project and your environment, it knows your config, your logs, your metrics and it's all secured to the IAM credentials of the user using Gemini Cloud Assist.”

Gemini Cloud Assist improves networking by guiding users through troubleshooting processes and narrowing down issues. Church demonstrated during the session how it can identify relevant trends, validate findings using raw data and pinpoint specific errors.

“We're very early in the Gemini Cloud Assist journey and we can already see a lot of promise,” Church said.

With an increasing volume of  artificial intelligence (AI) workloads running in the cloud, there is also a need to re-architect how networking works to optimize those workloads.

Anna Berenberg, engineering fellow at Google Cloud detailed multiple innovations to help improve networking for genAI application deployments.

“An AI application is not a web application,” Berenberg said. “While both have traffic-management goals of optimization for maintenance and efficiency, they're actually very different.”

She said that AI applications generally are more deterministic in terms of processing time which is quite different from a typical web application. Among the ways that Google Cloud is looking to address the challenge is with the debut of model-as-a-service endpoints, which treat inference as a service. Additionally, Google has developed optimized traffic management features for AI workloads, such as auto-routing, multi-modal affinity, and programmability with service extensions.

Looking ahead, Berenberg, believes that networking products will evolve to become AI-assisted, with capabilities like natural language support, intent-based configuration, and optimization across entire microservice stacks.

“We believe that networking products will become AI-assisted products and every product will have natural language support,” she said.