Arista SwitchApp is a full-featured 1/10/40G* Layer 2 switching app implemented in FPGA, optimized for low latency and integrated with the Arista EOS® (Extensible Operating System).


Arista SwitchApp

It delivers the convenience and simplicity of a traditional switch with the optimised latency and programmability of an FPGA-based system. SwitchApp runs on Arista's 7130LB devices, implementing a full Layer 2 switching pipeline for up to 48 ports. Rather than requiring a one-size-fits-all switch architecture, as would be necessary in fixed silicon, Arista's logic engineers have optimised several system profiles, each specifically tuned for the intended use case. The flexibility and programmability of FPGA fabric allows customers to obtain the best possible performance e.g. a simple L2 switch may achieve a lower latency than one that has more features. A switch with a low bi-section bandwidth achieves a lower latency than one with non-blocking bandwidth.

SwitchApp leverages the world's best network operating system – EOS – to provide access to a rich, and growing, set of features. As new features are added, the hardware logic can be updated, enabling entirely new use cases via a simple software update.

Featured Video: Arista 7130 SwitchApp Overview

 
Features Benefits
Low latency Forward packets at an ultra low latency of 94-132 ns** from any port to any other port. SwitchApp has been architected with latency as a first class design goal.
Bandwidth Use the non-blocking bandwidth profiles to provide up to 480 Gbps between ports.
Full L2 Switch Pipeline Implement a full-featured L2 switch pipeline, allowing for VLAN, Multicast, ACLs*, port aggregation and MLAG*. In use cases that don't require advanced L2 features, take advantage of a streamlined L2 profile to achieve the absolute lowest latency.
Full latency measurements are available for each build and each feature profile.
EOS Integration Take advantage of the familiar EOS configuration experience and feature set. Protocols such as RSTP, LLDP and IGMP work as per other traditional L2 switches. CLI, telemetry, counters, configuration automation and more are equally available.
CloudVision Integration Extend the EOS-based switch to a network-wide model for provisioning, orchestration, and telemetry. CloudVision moreover acts as a unified control point for third party overlay controllers, orchestration systems, and security platforms.
Scale Use 10,000 Unicast MAC addresses or Multicast groups. Future updates will support:
  • Up to 2,000 L3 ACLs per port*
  • 38,000 L3 routes*
  • 4,000 NAT flows*
FPGA-based Leverage the re-programmability of FPGAs to optimise switching architectures to support different use cases. New feature requests, normally reserved for hardware updates, can be implemented via software update augmenting the functionality and extending the life of existing investments.

Future improvements, delivered via software update, will include Layer 3 routing capabilities* (for example BGP, PIM, OSPF, VRRP), NAT* and IEEE 1588* (PTP boundary clock).
L1-Integration Keep the benefits of Layer 1 switching, alongside SwitchApp. 7130 devices which support both SwitchApp and L1 switching can do both simultaneously. Use L1 switching to implement tapping, low latency distribution and media conversion, all converged in the same device as the L2 switch.
Agile Ports Adapt from 1G to 10G and 40G without costly upgrades.
Deterministic Know and rely upon your system's latency for fairness, or implement an ideal execution environment for your orders. Full cut-through architecture in all profiles results in no latency variation with different length packets.

Optimized for:

  • Arista 7130LB devices with embedded Xilinx Ultrascale+ FPGA.

Use-Cases:

  • Multi-layer MLAG-based leaf-spine fabric, including redundant connections.
  • Exchange-facing switch with multiple exchange connections, L1 tapping, and redundancy.
  • Exchange-facing connectivity.
  • L2 Aggregation for latency sensitive medium-large Colo.
  • L2 Multicast pub/sub.
* Roadmap features
** Average latency quoted
Arista SwitchApp
Table 1.1 SwitchApp measures latency for Layer 2-Low-Latency and Layer 1-Full All Latency numbers quoted are 10G to 10G.