Success Story

HPE Juniper Networking Builds Next-Gen AI Infrastructure

To scale AI-ready networks, HPE Juniper Networking needed to validate data center switching performance against real-world AI workload demands before deployment. With Keysight, HPE Juniper Networking tested and optimized AI-ready infrastructure using AI-native emulation and validation tools.

test

Validating AI Network Performance at Scale

HPE Juniper Networking Case Study Highlights

Network Infrastructure Leader

HPE Juniper Networking designs high-performance networking platforms for some of the world’s most demanding data center environments. As AI workloads drive large-scale GPU deployments, its customers are placing increasing demands on network performance, scale, and reliability. These shifts make HPE Juniper Networking’s data center infrastructure a critical factor in GPU utilization, performance, and cost efficiency, raising the stakes for validating AI-ready architectures before deployment.

Proving Deployment Performance

HPE Juniper Networking needed to validate that high-radix QFX5230 and QFX5240 switching platforms could perform reliably under real-world AI traffic conditions, before customer deployment. AI environments generate long-lived, high-bandwidth traffic flows, known as elephant flows, that stress network architectures with East-West traffic, unlike traditional data center traffic. Emulation allows testing the fundamental behavior of features and validating command-line interface (CLI) knob mechanics. The challenge was proving system-level performance under these AI-specific conditions without building massive physical GPU clusters.

AI-Native Traffic Emulation

Using Keysight AresONE, a traffic emulation platform purpose-built for AI data centers, HPE Juniper Networking was able to emulate RoCEv2 traffic at line rates up to 800GE. Traffic can be bursty and exhibit specific patterns, such as all-to-all, with many GPUs trying to send data to one another, causing in-cast congestion. By recreating AI-specific traffic patterns and GPU cluster behavior, HPE Juniper Networking could model real-world congestion, buffering, and load-balancing scenarios. The Keysight solution emulates a GPU and remote direct memory access (RDMA) network interface card (NIC) to create realistic AI workload traffic. This approach enabled HPE Juniper Networking to validate advanced switch features and operational controls, without requiring large physical GPU deployments, reducing risk and accelerating readiness for customer environments.

Proven Performance at AI Scale

Using Keysight’s AI-native traffic emulation, HPE Juniper Networking validated and optimized the real-world performance of its high-radix switches under AI data center operating conditions. This enabled HPE Juniper Networking to accelerate feature validation and significantly reduce deployment risk. As a result, HPE Juniper Networking demonstrated scalable, reliable network performance, from hundreds to tens of thousands of GPUs.
Mahesh Subramaniam, Director of Product Management, AI Data Centers, Juniper Networks

“Our partnership with Keysight helps us deliver trusted networking infrastructure ready for today, and built for tomorrow’s AI challenges.”

Mahesh Subramaniam, Director of Product Management, AI Data Centers, HPE Juniper Networking

Explore the Tools HPE Juniper Networking Used

AresONE-M 800GE

Validate equipment interoperability and performance.

Keysight AI Data Center Builder

Design and deploy AI network infrastructure​ and emulate AI workloads.

Explore Related Resources

White Paper

White Paper

Differences Between AI / ML Networks and Traditional Networks

Top technical questions and answers on how AI/ML really impact data center networking and design.

White Paper

White Paper

Scaling AI Data Centers for Real-World Performance

Technical insight on emulation, validation, and cluster behavior at AI infrastructure scale.

Report

Report

AI Cluster Networking Bottlenecks, Scale, and What Comes Next

Survey-based insights on infrastructure investment, emulation gaps, interconnect adoption, and optimization priorities in AI data centers.

×

Please have a salesperson contact me.

*Indicates required field

Select a preferred method of communication*Required Field
Preferred method of communication? Change email?
Preferred method of communication?

By continuing, you are providing Keysight with your personal data. See the Keysight Privacy Statement for information on how we use this data.

Thank you

A sales representative will contact you soon.