Just as cloud computing led to the emergence of software-defined (SD) load balancing, the artificial intelligence (AI) revolution is taking us a step farther, to AI-defined architectures. This transformation represents a significant shift in how enterprises approach their infrastructure to support modern AI workloads and bring AI benefits to existing workloads.
AI applications present significant challenges with respect to load balancing. AI workloads, including agentic workloads, demand extreme performance: terabits/second, not the gigabits/second that’s been required for traditional applications. As a result, organizations need load balancers with extraordinary throughput capabilities and the scalability to support elastic operations.
“When you build modern AI applications for enterprises, there has to be a very high level of performance, resilience, security, and elasticity,” says Chris Wolf, global head of AI and advanced services, VCF Division at Broadcom. “Load balancers in the AI era must be able to manage services and fulfill enterprise requirements across multiple servers and clusters, because of the distributed nature of large inference and training jobs in private AI environments.”
Additionally, enterprise AI applications are almost exclusively built on Kubernetes with a microservices architecture. That means organizations need load balancers that can autoscale, autoheal, and operate “as code,” with built-in capabilities including global server load balancing (GSLB), web application firewalls (WAFs), and application programming interface (API) security.
AI applications exchange vast amounts of sensitive data through APIs, requiring robust protection against attacks and data leakage through comprehensive web app and API security. Thresholding with anomaly detection and traffic pattern recognition should be employed to optimize resource allocation.
AI-defined load balancing
It’s only fitting that load balancing in the AI era employs AI to get the job done, and it does so across three key dimensions.
First, predictive intelligence enables high resilience, by leveraging health score monitoring and dynamic thresholds that scale in real time as needed to accommodate bursts. In this environment, static thresholds aren’t feasible, because traffic is too dynamic and overprovisioning for max load would be prohibitively expensive. Active-active high-availability configurations ensure continuous operation, and autoscaling capabilities coupled with autohealing recognize traffic patterns and remediate most issues without an admin getting deeply involved, if at all.
Second, generative AI (genAI) can dramatically improve operational efficiency by acting as copilots to assist teams in several ways. Admins can ask questions by using natural language, and the AI tools provide answers, analytics, and contextual insights based on information found in application health scores, application latency measurements, design guides, and knowledge base (KB) documentation. AI tools can also provide correlated analytics, contextual insights, and multifactor inference within admins’ work streams. Infrastructure-as-code capabilities reduce manual work, because configurations can be changed programmatically in software. Capacity management and performance troubleshooting assistance can flag emerging issues for admins to address long before they affect users, all of which dramatically improves productivity.
Finally, AI-powered self-service capabilities create load balancing interfaces for DevOps teams that require zero training, because AI can provide intuitive assistance for engineers to follow. The result is faster deployment and configuration without sacrificing quality or security.
A solution that meets all of these AI era requirements, such as Broadcom’s VMware Avi Load Balancer, delivers big dividends. Rigorous studies have shown that enterprise IT can achieve 43% OpEx savings, 90% faster app delivery provisioning, and a 27% DevOps productivity boost with this solution.
Software-defined load balancing principles remain—ensuring scale-out performance, dynamic availability, and application-level security—and the AI era dramatically amplifies these requirements while infusing AI principles. Organizations that embrace AI-defined load balancing will not only support their AI and non-AI workloads more effectively but will also benefit from the intelligence embedded within their infrastructure.
To learn more about how Broadcom can help your organization bring load balancing into the AI era, visit us here.
>
>About the author:
>Umesh Mahajan is Vice President and General Manager of Broadcom’s Application Networking and Security Division. He joined Broadcom from VMware, where he led the Networking and Security Business Unit and was responsible for the NSX software-defined network virtualization platform, which encompassed network connectivity, security, and load balancing. With more than three decades of experience in multi-cloud networking and networking services, Mr. Mahajan holds over 30 patents. Prior to joining VMware, he founded Avi Networks, which built the disruptive software-defined advanced load balancer. Earlier, he held senior leadership positions at Cisco, including Vice President and General Manager of the data center switching business, and was responsible for Nexus 7000 & MDS 9000 platforms, and the NX-OS operating system. Mr. Mahajan holds a Master of Science in computer science from Duke University and a Bachelor of Technology from IIT Delhi.
LinkedIn: https://www.linkedin.com/in/umeshmahajan/
>
Read More from This Article: AI and load balancing
Source: News