7050X3 Series | Platform Overview

Dynamic Buffer Allocation

In cut-through mode, the Arista 7050X3 switches forward packets with a consistent low latency of 800 nanoseconds. Upon congestion,

the packets are buffered in an intelligent fully shared packet memory that has a total size of 32MB for superior burst absorption. Unlike

other architectures that have fixed per-port packet memory, the 7050X3 Series use dynamic thresholds to allocate packet memory

based on traffic class, queue depth and quality of service policy ensuring a fair allocation to all ports of both lossy and lossless classes.

Buffer utilization, occupancy and thresholds are all visible with Arista LANZ and can be exported to monitoring tools to identify

hotspots and measure latency at the device and end to end.

Software Driven Cloud Networking

Arista Software Driven Cloud Networking (SDCN), combines the principles that have made cloud computing the unstoppable force

that it is: automation, self service provisioning, and linear scaling of both performance and economics coupled with the trend in

Software Defined Networking that delivers: network virtualization, custom programmability, simplified architectures, and lower capital

expenditure. This combination creates a best-in-class software foundation for maximizing the value of the network to both the

enterprise and service provider data center. A new architecture for the most mission-critical location within the IT infrastructure that

simplifies management and provisioning, speeds up service delivery, lowers costs and creates opportunities for competitive

differentiation, while putting control and visibility back in the hands of the network and systems administrators.

Maximum Flexibility for Scale Out Network Designs

Scale out network designs enable solutions to start small and evolve over time. A simple two-way design can grow as far as 128-way

without significant changes to the architecture. The Arista 7050X3 include enhancements for flexible scale-out designs:

• 128-way ECMP and 64-way MLAG to provide scalable designs and balance traffic evenly across large scale 2 tier leaf-spine designs

• Equal and Unequal Cost Multi-Pathing (UCMP) for flexible traffic balancing in large scale multi-tier topologies

• Custom hash algorithms for efficient hashing, persistent hashing and custom lookups for tunneled protocols

• Flexible allocation of L2 and L3 forwarding table resources for more design choice

• Wide choice of dense 10G/25G/40G/50G/100G interfaces for multi-speed flexibility

• Support for standards based IEEE 25GbE for simple and cost effective migration from 10G and 40G to 25G and 100G

• VXLAN routing, bridging and gateway capability for physical to virtualization communication in next generation data center designs

• DANZ, sFlow and multi-port mirroring to detect micro-burst congestion and provide network wide visibility and monitoring

• Hitless speed changes from 10G to 100G to eliminate down-time when implementing speed changes

Unified Forwarding Table

Network scalability is directly impacted by the size of a switches forwarding tables. In many systems a 'one size fits all' approach is

adopted using discrete fixed size tables for each of the common types of forwarding entry. The Arista 7050X3 leverage a common

Unified Forwarding Table (UFT) for the L2 MAC, L3 Routing, L3 Host and IP Multicast forwarding entries, which can be partitioned per

entry type. The ideal size of each partition varies depending on the network deployment scenario. The flexibility of the UFT coupled

with the range of pre-defined profiles available on the 7050X3 ensures optimal resource allocation for all network topologies and

network virtualization technologies.

AI Analyzer *

Traditional software-based traffic counters do not lend themselves to examine AI/ML traffic patterns, which exhibit unique ramp up

behavior in very short intervals of time. The AI Analyzer is a hardware capability that enables the collection of ECMP member utilization

data, aggregated over extremely short periods of time. This allows the Arista 7050X3 series to effectively analyze the traffic patterns,

with a time interval as granular as 100 microseconds. The results of such an analysis can then be applied to fine tune dynamic load

balancing workloads uniformly across the ECMP member links, which is a key requirement for AI/ML applications.

CloudVision

CloudVision is a network-wide approach for workload orchestration and workflow automation as a turnkey solution for Cloud

Networking. CloudVision extends the EOS publish subscribe architectural approach across the network for state, topology, monitoring

and visibility. This enables enterprises to move to cloud-class automation without needing any significant internal development.

* Not currently supported in EOS