7700R4 ARISTA | Alldatasheet

Document overview

  • Manufacturer or author: Provided By alldatasheet.com(free datasheet download site)
  • PDF pages: 7

Technical content

7700R4 Series Distributed Etherlink Switch Data Sheet Product Highlights System Scale and Performance

  • Over 27k x 800Gbps Interfaces
  • 22 Pbps (44 Pbps FDX) total capacity
  • Up to 8 Trillion packets per second
  • Up to 24 TB of system-wide buffer Optimized for AI/ML and HPC Clusters
  • Fully scheduled lossless interconnect
  • 100% efficient cell based load balancing
  • Distributed hardware scheduler
  • Per port Virtual Output Queuing to eliminate head of line blocking
  • Hardware accelerated link health management
  • Under 4 microsecond latency (64 bytes)
  • Ultra Ethernet ready Arista Etherlink for AI
  • AI Analyzer powered by AVA
  • Advanced collective load balancing and congestion management
  • AI workflow integration High Availability
  • Hot swappable redundant power
  • Field serviceable supervisors
  • Integrated over provisioning
  • Real-time hardware health checking
  • Graceful uplink capacity degradation Power Efficiency
  • High efficiency power and cooling design
  • ORv3 54V power support
  • High performance SerDes enable
  • 4m passive copper (DAC)
  • Linear-drive pluggable optics (LPO) Resilient Distributed Control Plane
  • Fully distributed architecture
  • No single point of failure
  • High performance 8-core x86 CPU Arista Extensible Operating System
  • Single 64-bit binary image
  • Fine-grained truly modular network OS
  • Stateful Fault Containment & Repair
  • Full access to Linux shell and tools
  • Open APIs and Real-time telemetry
  • Extensible platform - bash, python, C++ Overview The Arista 7700R4 Distributed Etherlink Switch (DES) is the first in a new series of ultra-scalable, smart distributed systems that take the architectural foundations of the 7800R beyond the con fines of a chassis to deliver thousands of 800GbE ports and Petabits per second of capacity combined with perfect fairness and lossless packet delivery - imperative for supporting large scale, open, AI and ML applications built on any flavor of XPU. The Ultra Ethernet ready DES system is designed exclusively to support the needs of leading edge, very large scale Ethernet based generative AI clusters which depend on the interconnect between accelerators for the e fficient execution of thousands of simultaneous high bandwidth transactions. Congestion, packet loss and failure can dramatically reduce the e fficiency of a workload or potentially lead to a stall in processing, all of which reduce the useful working time and therefore return on investment (ROI) of high value compute farms. The unique scalability and fully co-ordinated logical single-hop behavior is made possible by an automated and fully hardware accelerated architectural framework. Massively parallel, distributed, end-to-end scheduling combined with deterministic traffic spraying across all available leaf-spine interfaces delivers 100% e fficient path utilization and equality to all tra ffic flows with no tuning required. These are critical attributes for supporting both single large workloads as well as mixed multi-tenant and multi-generational jobs in parallel. As a distributed system, DES is designed for pay-as-you-grow scaling starting from hundreds of ports, with a predictable, linear CapEx and OpEx trajectory to maximum capacity. Arista Etherlink for AI adds layers of smart, workload speci fic functionality improving, tra ffic management, con figuration consistency and visibility - critical factors in building, operating and troubleshooting very large clusters and minimizing downtime. Arista EOS All Arista products including the 7700R Series run the same Arista EOS software, simplifying network administration with a common standard across all switches. Arista EOS is a modular switch operating system with a unique state sharing architecture that cleanly separates switch state from protocol processing and application logic. Built on top of a standard Linux kernel, all EOS processes run in their own protected memory space and exchange state through an in-memory database. This multi-process state sharing architecture provides the foundation for in-service-software updates and self-healing resiliency together with stateful switchover without the loss of data plane forwarding. Arista EOS enables advanced monitoring and automation capabilities such as AI Analyzer, LANZ, Zero Touch Provisioning, and enables Linux based tools to be run natively on the switch. Arista 7700R Distributed Etherlink Switch

The 7700R4 Distributed Etherlink Switch series builds on seven previous generations of Arista 7500 and 7800 series systems, leveraging the fundamental system architecture that is proven by hundreds of thousands of systems in the world’s largest and most critical networks and is acknowledged as the most efficient solution for scaling back-end networks for AI. Where the 7500 and 7800 Series offer modularity within a single system, the 7700R series enhances scaling beyond single systems, enabling scalability to thousands of ports. Separating the traditional line card and fabric functions into discrete devices enables higher scalability through substantially larger topologies than is possible in a single physical chassis while also distributing power consumption, cooling needs and the control plane across multiple devices, and enabling hosts to connect to distributed leaves using cost effective, low power passive copper cabling. The fully distributed control-plane runs locally on each system and combines with hardware based topology con figuration, integrated end to end scheduling, distributed queueing and automatic link health monitoring to minimize deployment time and maximize performance and reliability in service. The 7700R4 system consists of two major components, which are deployed in similar ways to a typical leaf-spine con figuration that creates single system topologies that scale to over 27,000 x 800GbE or over 31,000 x 400GbE host ports.

  • 7700R4C-38PE - 38 Port 800GbE leaf switch with 18 host ports and 20 fabric ports
  • 7720R4-128PE - 128 Port 800GbE spine switch with 128 fabric ports Arista 7700R4 systems can be combined in various topologies to provide a choice of density, scale and speed, with linear CapEx and OpEx. 7700R4 Series systems are designed with maximum e fficiency in mind, based on an advanced 5nm silicon geometry combined with optimized power and cooling design. Support for passive copper cables up to 4m and Linear-drive Pluggable Optics dramatically reduces the power and cooling footprint of very large scale networks. 7700R4 Series | Platform Overview 7700R4C-38PE: 38 port 800GbE Distributed Leaf Switch - 18 x 800GbE host facing ports - 20 x 800GbE spine facing fabric ports - 14.4 Tbps of forwarding and 5.4 Bpps with 16GB of buffer - Field serviceable 8-core x86 supervisor 7700R4-128PE: 128 port 800GbE Distributed Spine Switch - 128 x 800GbE leaf facing fabric ports - 102.4 Tbps non-blocking fabric bandwidth - Field serviceable 8-core x86 supervisor Rear view of 7700 Series systems - 7700R4C-38PE is shown with F-R airflow and PWR-2411-MC - 7720R4-128PE is shown with F-R airflow and PWR-2421-HV

7700 DES Fully Scheduled Lossless Network Performance

The Arista 7700R Series Distributed Etherlink Switch is optimized to deliver massively parallel end to end scheduling and coordination - even in the largest topologies, the platform operates like a single extremely large switching chip that is 100% internally lossless and fair. Four advanced technologies converge to this highly desirable architecture:

  • Cell-based traffic spraying fabric - elegantly mitigates elephant and mice flows. Uniform traffic spraying for 100% efficiency
  • Virtual Output Queues (VOQ) - invokes ingress virtual queues to every egress port, eliminating head of line blocking (HOLB)
  • Distributed Credit Scheduling - ensures all egress ports are independent eliminating HOLB and noisy neighbors
  • Deep Buffering - easily handles in-cast, bursts and speed mismatches without packet loss, keeping TCP running efficiently This advanced architecture enables the Arista 7700R4 to handle the most demanding cluster computing workloads with ease. Generative AI clusters, ML and HPC environments all bene fit from the 7700R4 Series’ ability to linearly scale bisectional bandwidth and logical radix while handling high bandwidth low entropy tra ffic and mixed tra ffic loads without increasing latency or introducing unnatural congestion. 7700R4 Series | Platform Overview 7700R4 Series Fully Scheduled Architecture

7700 DES Topology and Workload Flexibility

AI training workloads typically involve synchronized, high throughput tra ffic flows that require careful network design to maximize XPU efficiency and performance. Building a non-blocking, fully-connected fabric is a common starting point but optimal design depends on a variety of factors. Traffic patterns and their appropriate topologies are deployment specific and may require specific connectivity plans which require specialized physical topologies. Planar (rail based) topologies, for example, require a unique physical cabling plan between accelerators and leaf switches which makes running mixed workloads on the same cluster more complicated. Factors that influence network design may include:

  • The number, type and behavior of workload(s) and model(s) in use
  • How workloads are sub-divided (e.g. sharding)
  • Number of accelerators per workload and how they are distributed across the cluster
  • The selection of accelerator processor architecture and the requirements for network connectivity
  • Optimizations provided by the Collective Communications Libraries (CCL) for each workload
  • Levels of resilience / redundancy required and tolerance to reduced performance DES’ innovative logical single-hop architecture, robust tra ffic management and inherently fair forwarding paradigm eliminates the complexity of workload-speci fic architectures and tuning, abstracting logical and physical topologies while enabling multiple unique workloads in parallel. In many cases, DES removes the need for unique, job speci fic, cabling plans, as every accelerator is logically connected to the same single distributed switch. This allows the use of low cost, short reach cables and optics, saving signi ficant expenditure and power consumption on cross-facility optical interconnections and eliminating the need to re-cable for di fferent workload topologies.

Deploying the Distributed Etherlink Switch A typical deployment of the 7700R4 Distributed Etherlink Switch follows the traditional, well understood and widely implemented, 2- tier leaf-spine physical topology model. Each leaf 7700R4 switch provides 18 x 800GbE ports for host connectivity and 20 x 800G fabric ports which are designed to be distributed evenly across a set of 7720R4 spines. Consistent with other leaf-spine designs, spines only connect to leaf switches, not to each other or to other networks. In a simple 2-tier configuration, DES scales to support over 4600 x 800GbE hosts (4x the size of the largest 7816R4 modular system), in a single logical system with single-hop logical forwarding. By introducing an orthogonal super-spine layer, DES can continue to expand to support tens of thousands of 400GbE or 800GbE connected accelerators. DES Leaf switches are built with an inherent over-provisioning ratio (18 host ports : 20 spine ports) which serves as an integrated redundancy mechanism to avoid congestion in the event of either an uplink or spine failure, allowing workloads to continue unaffected. It is typical to connect up to 16 accelerators to each leaf switch with the remaining two ports per leaf being used for storage nodes or inter-cluster connectivity. Starting with the total number of accelerator ports required, the number of leaf switches can be determined and subsequently the number of spine switches necessary to support equal connectivity to all leaf devices. In typical deployments, each leaf would be Despite physically resembling a two or three tier leaf-spine network, DES deployed at any scale continues to provide fully co-ordinated, logical single-hop forwarding, eliminating the traditional congestion management protocols and tuning required by a leaf-spine constructed of independent devices. With fully automated fabric con figuration and operation, overheads and time to operation are minimized while providing deterministic performance for any workload. 7700R4 Series | Platform Overview 7700R4 Series Typical Deployment Topology Scaling Examples 1152 x 800G Hosts 2304 x 800G Hosts 4608 x 800G Hosts Over 4608 x 800G Hosts Number of Spine Nodes 10 20 40 Up to 800 inc. Super Spine Number of Leaf Nodes 64 128 256 Over 1500 Leaf-Spine Interconnect 16 Tbps (1600G per spine) 16 Tbps (800G per Spine) 16 Tbps (400G per Spine) 16 Tbps (400G per Spine)

Model Comparison 7700R4C-38PE 7720R4-128PE 800G Host Ports 18 x OSFP800 — 800G Fabric Ports 20 x OSFP800 128 x OSFP800 Additional Ports 1 x QSFP100 — Throughput (FDX) 14.4 (28.8) Tbps 102.4 (204.8) Tbps Packets/Second 5.4 Bpps — End to End Latency From 3.8 us CPU Eight-Core x86 Eight-Core x86 System Memory 64 Gigabytes 64 Gigabytes Packet Buffer Memory 16 GB — Supervisor Module 7001-SUP-EL 7003-SUP-EL USB and RS232 Ports 1 USB, 1 RS232 (RJ45) 100/1000 Mgmt Ports 1 2 Hot-swap Power Supplies 2 (1+1 redundant) 4 (N+N redundant) Hot-swappable Fans 4 (N+1 redundant) 12 (N+1 redundant) Airflow Direction (Fan Tray) Front to Rear Rack Units 2 U 7 U Size (WxHxD) inc. handles 17.32 x 3.46 x 23.06 in ( 44 x 8.8 x 58.6 cm) 17.32 x 12.11 x 22.06 in ( 44 x 30.8 x 56 cm) Typical/Max Power Draw 1 TBD TBD Weight 47.2 lbs (21.4 kg) 110.3 lbs (50 kg) Fan Tray FAN-7021H Power Supplies PWR-2421-HV or PWR-2411-MC EOS Feature Licenses LIC-DL — Minimum EOS TBD 1. Typical power consumption measured at 25C ambient with 50% load on all ports. Excludes optics power as this is a significant variable. 7700R4 Series | Technical Specifications

7700R4 Series |Technical Specifications Supervisor Module 7001-SUP-EL 7003-SUP-EL Processor Eight-Core x86 Eight-Core x86 System Memory 64 GB 64 GB SSD Storage 512 GB 512 GB Default Secure Boot Yes Yes Size (WxHxD) TBD TBD Weight TBD TBD Typical (Max) Power 1 TBD TBD System Compatibility DCS-DL-7700 DCS-DS-7720 Power Supply Specifications Power Supply PWR-2421 HV PWR-2411 MC Input Voltage 200-277VAC 240-380VDC 48 to 72 VDC Max Input Current 13.5 A (200V AC) 70 - 40 A Max ( 43 - 72 V) Input Frequency 50/60 Hz or DC DC Output Power 2400 W 2400 W Input Connector SAF-D 6-Pin M-CRPS Efficiency (Typical) 96% 96% Standards Compliance EMC FCC Class A, ICES-003, EN 55032, EN IEC 61000-3-2:2019, EN 61000-3-3 Immunity EN 55035 EN 300 386 Safety EN 62368-1:2014 + A11:2017 IEC 62368-1:2014 Certifications BSMI (Taiwan) CE (European Union) KCC (South Korea) NRTL (North America) RCM (Australia/New Zealand) UKCA (United Kingdom) VCCI (Japan) European Union Directives 2014/35/EU Low Voltage Directive 2014/30/EU EMC Directive 2012/19/EU WEEE Directive 2011/65/EU RoHS Directive 2015/863/EU Commission Delegated Directive Further Information Product Certification Portal Environmental Characteristics Operating Temperature 0 to 40˚C (32 to 104˚F) Storage Temperature -40 to 70˚C (-40 to 158˚F) Relative Humidity 5 to 95% Operating Altitude 0 to 10,000 ft, (0-3,000m) 1. Typical power consumption measured at 25C ambient with 50% load on all ports. Excludes optics power as this is a significant variable.

7700R4 Series | Ordering Information Headquarters

5453 Great America Parkway

Santa Clara, California 95054 408-547-5500 Support support@arista.com 408-547-5502 866-476-0000 Sales sales@arista.com 408-547-5501 866-497-0000 Copyright 2024 Arista Networks, Inc. The information contained herein is subject to change without notice. Arista, the Arista logo and EOS are trademarks of Arista Networks. Other product or service names may be trademarks or service marks of others. Warranty The Arista 7700 Series switches come with a one-year limited hardware warranty, which covers parts, repair, or replacement with a 10 business day turn-around after the unit is received. Service and Support Support services including next business day and 4-hour advance hardware replacement are available. For service depot locations, please see: http://www.arista.com/en/service www.arista.com September 2nd, 2024 03-0077-01 Systems and Optional Components DCS-DL-7700R4C-38PE-F 7700R4 Series Distributed Leaf Switch with 18 x 800G OSFP ports and 20 x 800G OSFP fabric ports, Sup-EL, front to rear air, 2 x AC and 2 x SAF-D cords DCS-DL-7700R4C-38PE# 7720R4 Series Distributed Spine Switch Bundle with 128 x 800G OSFP fabric ports, Sup-EL (no fans or PSU) DCS-DS-7720R4-128PE-F 7720R4 Series Distributed Spine Switch Bundle with 128 x 800G OSFP fabric ports, Sup-EL, front to rear air 4 x AC and 4 x SAF-D cords DCS-DS-7720R4-128PE# 7720R4 Series Distributed Spine Switch Bundle with 128 x 800G OSFP fabric ports, Sup-EL (no fans or PSU) DCS-DS-7720-CH Arista 7720 chassis, 1 Supervisor slot, 1 Switch Card slot, AC or DC option DCS-7001-SUP-EL 7001 series Supervisor-EL module with 8c/16t 2.3 GHz CPU, 64GB RAM, 512GB NVMe (spare, -F airflow, secure boot enabled) DCS-7003-SUP-EL 7003 series Supervisor-EL module with 8c/16t 2.3 GHz CPU, 64GB RAM, 512GB NVMe (spare, -F airflow, secure boot enabled) FAN-7021H-RED Spare Front-to-Rear High Speed Fan module for select Arista 7000 Series. Refer to product datasheets for compatibility. FAN-7021H-RED# Configurable Front-to-Rear High Speed Fan module for select Arista 7000 Series. Refer to product datasheets for compatibility. FAN-7021H-BLUE Spare Rear-to-Front High Speed Fan module for select Arista 7000 Series. Refer to product datasheets for compatibility. FAN-7021H-BLUE# Configurable Rear-to-Front High Speed Fan module for select Arista 7000 Series. Refer to product datasheets for compatibility. PWR-2411-HV-RED Spare Arista PSU, 1RU, HVAC/DC, 2400W, SAF-D, FORWARD, 73.5MM PWR-2411-HV-RED# Configurable Arista PSU, 1RU, HVAC/DC, 2400W, SAF-D, FORWARD, 73.5MM PWR-2411-MC-RED Spare Arista PSU, 1RU, +48VDC, 2400W, 6-PIN Modular Connector, FORWARD, 73.5MM PWR-2411-MC-RED# Configurable Arista PSU, 1RU, +48VDC, 2400W, 6-PIN Modular Connector, FORWARD, 73.5MM