Innovium’s TERALYNX Switches: Addressing a Range of Deployment Use Cases across Public & Private Cloud

We recently introduced TERALYNX, the world’s fastest family of data center Ethernet switches with up to 12.8Tbps performance, featuring line-rate programmability, super-low latency, comprehensive data center tunneling, optimized on-chip buffers along with unmatched telemetry capabilities. The customer reception has been outstanding and we look forward to supporting a broad range of public and private cloud use cases with our flagship product line.

Before we look at specific deployment scenarios, it is important to understand critical market drivers. Data center customers are seeing an explosion of network traffic, which is a result of many factors:

  • IP Storage: move to IP-based storage, adoption of hyper-converged storage, rise of flash storage and move to NVMe over fabrics
  • New Application Architectures: move to distributed, micro-services-based applications using containers and virtual machines drive east-west traffic significantly
  • Data Analytics: need for data analytics to drive business results requires low latencies and high performance
  • Machine Learning/AI: customers are beginning to use large machine learning/artificial intelligence clusters to recognize and process patterns from massive amounts of data that require the lowest latencies and highest performance
  • Images & Video: Increasing use of high resolution images and video

Some large data center operators have publicly stated that networking demand is one of the fastest growth area in their infrastructure and they are seeing doubling of traffic every 12-15 months. In addition, they are looking for capabilities like:

  • Real-time telemetry data to do automation and analytics, thereby enabling an autonomous network
  • No-compromise programmable networks to future proof it by supporting new protocols (e.g. tunneling type) and connectivity rates (e.g. 200G & 400G), while preventing unsatisfactory trade-offs on latencies and power, etc.
  • Essential tunneling capabilities such as VXLAN and IPinIP
  • Large buffers with optimized architectures along with best-in-class latencies to deliver superior network quality and fewer packet drops
  • Highly resilient and power efficient networks with compelling cost of ownership

Innovium TERALYNX is the perfect solution for these requirements, as it enables unmatched, scalable, high-performance, programmable networks. Innovium TERALYNX offers a broad, software compatible product portfolio of 12.8Tbps, 6.4Tbps and 3.2Tbps, ideal for all tiers of a large data center network. Innovium offers a game-changing advantage in performance – a single Innovium TERALYNX 12.8Tbps switch can replace twelve 3.2Tbps switches or six 6.4Tbps switches, thereby dramatically cutting down cost, complexity, power, and latency as well as simplifying software and management. Further, it delivers huge advantages with superior buffering and latency to provide significantly improved network performance and quality, programmability, tunneling and advanced telemetry which enables rapid troubleshooting and automation.

Use cases for Innovium TERALYNX switch ASIC span both public and private cloud data-centers. Here are a few important ones:

    Click to enlarge
  • 25/50/100G Server connectivity in Cloud & Enterprise Data Centers: 25/40G NICs are mainstream already in large data centers. Customers are looking to move to 50/100/200G NICs next. As they do that, TERALYNX can be deployed in all tiers of the network to support the next-generation of bandwidth needs. Customers deploying a 1RU, 32 x QSFP switch today will likely deploy a 1RU, 32 x QSFP-DD/OSFP switch in the future to enable connectivity configurations such as 128 x 100Gbps, 64 x 200Gbps, or 32 x 400Gbps.
  • Deploying High Radix Fixed Switches in all tiers of the Enterprise & Cloud Network: Historically, as performance and radix have increased in switch silicon, fixed switch adoption has also increased. Fixed switches offer significant benefits in terms of cost, complexity and power over equivalent modular switches. Today, a fixed TOR switch is the default form-factor in the access layer. With TERALYNX 12.8Tbps switch silicon, customers can build a fixed switch solution with 128 ports of 100Gbps, 64 ports of 200Gbps or 32 ports of 400Gbps. We expect these fixed switches to be deployed in higher tiers (leaf & spine) of the network where modular switches are common today.
  • Flatten the network with larger radix in the Enterprise and Cloud: Today, a 3-tier network is quite common in the data center. It consists of access (TOR), aggregation (leaf) and core (spine) layers. With a radix of 128 ports of 100Gbps in the TERALYNX switch, customers can flatten the network to 2-tiers in a mid-size data center. This helps simplify the network by flattening it and reduces expensive optics connectivity, hops/latency, and overall system costs.
  • Newer topologies beyond standard TOR-Leaf-Spine network designs: Today’s standard data center network topology is that servers connect to a TOR, which then connects using ECMP to a Leaf-Spine network core. With TERALYNX’s large radix (128 x 100Gbps), customers are pursuing new topologies where TORs can further be connected to each other in a mesh. This helps reduce hops, latencies and congestion for inter-server connectivity, thereby enabling higher performance applications.
  • Low-latency AI/machine learning and HPC clusters: High performance applications want highly scalable, high performance and low-latency Ethernet networks. Machine learning and artificial intelligence applications demand similar capabilities. TERALYNX offers best in-class latency of less than 350 nanosecond, large radix capability and performance of up to 12.8Tbps, which is ideal for those deployments.
  • Low-latency, lossless Ethernet fabric for Flash/NVMe Storage in Cloud & Enterprise: Flash based IP storage deployments have been growing rapidly. NVMe based storage over Ethernet fabric is expected to ramp as well. Hence, customers are looking for higher performance, low-latencies, bigger buffers and RoCEv2 capabilities in the network fabric. TERALYNX is the perfect solution for those customers.
  • Using PAM4 SerDes and 50/100G lambda optics for 2x+ cost/bit savings: Cost per bit is a very important metric for large data center customers as they continue to expand networking capacity. 50G PAM4 Ethernet signaling and 50/100G optical lambdas (e.g. DR4, FR4, FR8) dramatically reduce cost per bit, by reducing SerDes lanes, optical components inside transceivers and optical fiber. As customers scale their network, they can use a TERALYNX PAM-4-based switch to scale their networks and reduce cost per bit.

For each of these use cases, customers can support future protocols or unique features using the programmability capability on TERALYNX to get long term investment protection. Unlike today, customers do not have to replace switches. Furthermore, advanced telemetry in TERALYNX enables rapid troubleshooting and automation necessary for autonomous networks. Finally, TERALYNX offers a highly power efficient solution that delivers customers much better performance/watt than switches today.

If you see a use case that matches your network deployment, or if you’d just like to discuss how TERALYNX might benefit your data center network and obtain additional details, please reach out to us at [email protected] We look forward to engaging with you to make your network higher quality, autonomous, more scalable and cost-effective than ever before!