Product Description
Broadcom Trident 5 / BCM78800 Series High-Capacity StrataXGS Trident 5 Programmable Ethernet Switch Series
The StrataXGS Trident 5 BCM78800 series is a family of Ethernet switch chips designed to address performance, capacity and service requirements for next-generation data center and cloud computing environments as well as high-bandwidth enterprise and mobile network applications. The Trident 5 architecture delivers compiler-driven, user-programmable forwarding and instrumentation capabilities at a very high port density, while maintaining minimum power, latency and board footprint. Trident 5 is designed to provide an optimal Top of Rack (ToR switch) solution.
Trident 5 introduces a new on-chip, neural-network inference engine called NetGNT (Networking General-purpose Neural-network Traffic-analyzer). NetGNT is an ML inference engine and can be trained to look for different types of traffic patterns that span the entire chip. For example, NetGNT can look for a traffic pattern common in AI/ML workloads, called “incast,” in which many packet flows all converge on the same port and buffer at roughly the same time, causing congestion. NetGNT can recognize this pattern in real-time as it begins, and invoke congestion-control techniques to avoid degraded network performance. And since NetGNT runs in hardware at full line rate, there is no impact to throughput or Ethernet switch latency.
Trident 5 has extensive features to address the rapidly increasing scale of data center network switch deployments and distributed computing applications such as running AI/ML workloads. These features include flexible in-band and streaming telemetry, large-scale forwarding databases with flexible allocations to allow tailoring to the required application, and support for the latest in datacenter protocols.
To enable high network utilization, Trident 5 features a wide variety of load balancing and congestion management technologies including Dynamic Load Balancing, Dynamic Group Multipathing, Resilient Hashing, Latency-Based ECN marking, and Elephant Flow detection/re-prioritization.
Trident 5 is programmed using the NPL high-level language. Customers can use off-the-shelf NPL applications from Broadcom or can license the NPL DIY environment and develop their own customized applications. For more information on NPL.
Features
Key attributes and benefits of the new BCM78800 series include:
- Robust connectivity using up to 160 instances of the industry’s best-performing, longest-reach 100G PAM-4 integrated SerDes core, enabling a wide variety of optical and direct attached copper (DAC) links with up to 4m DAC reach
- Cost- and power-optimized for next-generation data center and campus networks, supporting 400G ToRs (eg 24x400G + 8x800G) with a compiler-programmable architecture
- Support for 800G Ethernet ports
- On-chip traffic-analysis neural-network engine, NetGNT, provides enhanced capabilities for telemetry, security and traffic engineering
- Increased parsing depth compared to the Trident 4 family, providing future-proofing for newer telemetry and tunnel encapsulation requirements in the fast-evolving datacenter markets
- An extremely rich, enterprise-grade feature set with a high degree of feature concurrency
- Architecture scalability for bandwidth requirements throughout the enterprise network: SKUs available up to 16.0Tbps, all supported with the same NOS and hardware code base
- Industry-leading fully shared packet buffer and database sizes
- Extensive, programmable in-band telemetry including support for IFA 2.0 (In-band Flow Analyzer version 2)
- A wide variety of load-balancing and congestion management features including Dynamic Load Balancing, Dynamic Group Multipathing, Resilient Hashing, Latency-Based ECN marking, and Elephant Flow detection and re-prioritization
- Broadview Gen 4 integrated network instrumentation feature set and software suite, providing full visibility to network operators into packet flow behavior, traffic management state and switch internal performance
- Six GHz-class on-chip processors, enabling powerful out-of-band streaming telemetry and a variety of Broadcom-provided embedded applications such as 1588 and BFD
- Implemented in power-efficient 5nm process technology
Applications
- Data center programmable ToR (Top-of-Rack)
- Data center leaf/spine
- Switch blade in a server chassis
- Line card and fabric in a switch chassis
- Border/Data Center Interconnect (DCI)
- Edge Compute data center disaggregation