Product Overview

The rapid progress of AI can put significant pressure on data centers, highlighting the importance of efficient scaling and optimization.

Scalable infrastructure is crucial for AI data centers to manage the increasing demands of AI workloads and evolving technology. Today’s AI data centers demand more than component-level validation. Interoperability, performance, and efficiency are system-level metrics that are only measurable under real-world network conditions. Comprehensive test strategies must incorporate both physical-layer testing and full-stack AI workload emulators capable of optimizing system-level and interconnect performance. 

By focusing on scalable infrastructure, AI data centers can achieve higher efficiency, reliability, and performance to support the rapid advancement of AI technologies. 

The Interconnect and Network Performance Tester 1600GE (INPT-1600GE) is a new 1.6 Terabit Ethernet (1.6T) hardware test platform for traffic generation and analysis. It provides functional, performance and interoperability benchmarks over a broad range of silicon chips, interconnects, cables, and networking equipment.

The INPT-1600GE, with the  Interconnect Test System software (ITS), delivers a holistic solution that organizes, stores, and uses data intelligently to automate interconnect validation in high-speed Ethernet networks and AI data centers. 

Keysight’s Interconnect Test System (ITS) software makes it easier to qualify and validate interconnect performance. ITS, an advanced, cohesive software application is designed to validate AI infrastructure, network components, and data center interconnects from 200GE to 1600GE, expanding on the capabilities of the Interconnect and Network Performance Tester 800GE benchtop (INPT-800GE). The ITS software runs on both the 800GE and 1600GE INPT hardware platforms.

Key Features

Deploy a versatile high-speed Ethernet platform to increase test plan coverage
    •  Select a benchtop unit or rackmount chassis.
       --  Benchtop Unit: Offers two OSFP1600 front panel ports in a portable, lightweight, and office-quiet form factor. This model enables manufacturers of 1.6T silicon chips, optical transceivers, and cables to validate interconnection media for data center and AI network infrastructures.
       --  Rackmount model: Supports regression test beds to validate released products as firmware and software updates are applied to the ASICs, switches and routers. This chassis is best suited for manufacturers of 1.6T switch/router ASICs and networking equipment to help them meet the extensive test requirements of AI data centers 
    •  Support 1x1600GE, 2x800GE, 4x400GE, and 8x200GE configurations.
       --  Rackmount model also supports high port count 1600GE test beds, up to 32x1600GE ports.
    •  Test reliability, stability, performance, and interoperability of Ethernet devices and interconnects that use 212 Gb/s electrical lane interfaces, including silicon chips, optical transceivers, active optical and copper cables, and networking equipment
    •  Facilitate high power consumption optical transceivers by delivering up to 40 watts of power and cooling per port.

Optimize Interconnect Performance
    •  Offers Keysight’s traditional Layer 1 Bit Error Rate Testing (BERT) and BERT Inferred FEC (BIF).
    •  Provides additional lower layer test capabilities:
       --  Physical Coding Sublayer (PCS)
       --  Forward Error Correction (FEC)
       --  Frame Loss Ratio (FLR) validations
        Ensuring interconnects are ready for production network deployment.
    •  Addresses lab test beds for Layer 2 and Layer 3 RFC benchmarks, network, and internetworking protocol emulation for scale and performance validation (with the rackmount chassis). 
    •  Uses extensive Common Management Interface Specification (CMIS) features to enable efficient read/write configuration and reporting tasks. The platform automatically detects Application Selection Code settings and monitors digital optical signals to ensure optimal performance and easier management.

Benchmark and Enhance Productivity in AI interconnects and network infrastructure
    •  Automate, realistically emulate, and validate AI and machine learning workloads.
    •  Provide fast, reliable, scalable bench marking and performance testing, enabling operators to gain deep insights into collective communication performance. 
    •  Utilize Keysight’s IxNetwork software with:
        --  RoCEv2 transport protocol
        --  Data Center Quantized Congestion Notification (DCQCN) congestion control
        --  Priority Flow Control (PFC)
        Provide a scalable and cost-effective solution to validate data plane traffic management effectiveness in AI clusters, and to optimize network fabric performance.
    •  Deliver a comprehensive set of  tools and measurements with the ITS software. ITS enables users to rapidly build and automate test suites to qualify and benchmark the performance of various 800GE and 1600GE high-speed interconnects.
    •  Reduce GPU wait time by lowering latency and increasing data throughput.
    •  Decrease training time and the time needed to create functional Large Language Models (LLMs).

Interconnect and Network Performance Tester feature summary

Provide critical measurements for high-speed interconnect tests
The ITS software supports all Layer 1, Layer 1.5, and Layer 2 measurements required to characterize and validate the performance of optical and copper interconnects from 200GE to 1600GE PAM4 Ethernet speeds. ITS capabilities include:
    1. BERT for traditional layer 1 electrical test
    2. BERT Inferred FEC (BIF) 
    3. FEC performance and margin on all electrical lanes
    4. New optics and copper cable qualification
    5. Validation of CMIS with CMIS Compare
    6. Validation of all Ethernet speeds
    7. Layer 2 Ethernet frame Tx and Rx traffic
    8. System level performance and interoperability
    9. RFC Benchmark performance and stress test in real systems

Test 1.6 Terabit Ethernet over 212 Gb/s electrical lanes As the IEEE, OIF, and LPO-MSA continue to define 224 Gb/s specs, Keysight is expanding our portfolio of solutions that can validate 212 Gb/s Ethernet devices, interconnects, and network equipment for production deployment. The 212 Gb/s PAM4 signaling is sensitive to noise, interference, and crosstalk, which makes thorough validation from silicon chips to routers critical. The Keysight Interconnect and Network Performance Tester 1600GE (INPT-1600GE) tests reliability, stability, performance, and interoperability of Ethernet devices and interconnects that use 212 Gb/s electrical lane interfaces, including silicon chips, optical transceivers, active / passive cables, and networking equipment. INPT-1600GE provides BERT, FEC, Ethernet traffic generation, and error injection for high-speed devices. It can test electrical, optical paths, error correction, and clock recovery at both discrete and system levels, including stress testing to the breaking point. High-speed connections require the best signal integrity, low BERs, reliable FEC performance, and multi-vendor interoperability.  The Keysight INPT-1600GE benchtop and half-rack systems meet this challenge, offering versatile and scalable Layer 1 to Layer 3 Ethernet performance validation in various form factors.

Validate AI Data Centers and Network Infrastructures AI / ML and High-Performance Computing (HPC) are driving rapid growth in the build-up and scale-out of new network infrastructures in large data centers. There are also many different types of optical transceivers, and active and passive interconnect devices, each with different wavelengths, media types, operating ranges, and multi-rate Ethernet speed support. Network operators use the Common Management Interface Specification (CMIS) to manage and configure interconnects. There is an increased need to validate the functionality and performance of various optical and copper interconnects in their configurations. Once connected to networking equipment, additional testing is required for multi-vendor interoperability, link quality, reliability, and performance benchmarking. The INPT-1600GE provides critical test and validation capabilities from 200GE to 1600GE over 212 Gb/s Ethernet interfaces. The platform, working with the ITS software helps increase productivity in AI networks by reducing GPU wait time and training time. The solution also decreases training time and the time needed to create functional Large Language Models (LLMs).