Switch Stacking: Arista will "SWAG" ins Campusnetz bringen

Contents

With the introduction of the Switch Aggregation Group (SWAG) function in the standardized switch operating system Arista EOS and the CloudVision Leaf Spine Stack (LSS), Arista aims to significantly simplify the management of campus networks. By using Ethernet instead of proprietary solutions, the aim is to increase flexibility and reduce complexity at the same time.

The aim of the switch specialists from Santa Clara is centralized switch management that scales from two devices to entire campus networks. Is this the big attack on Cisco and HPE Aruba in the campus switching market?

Switch grouping with Ethernet

With the new Switch Aggregation Group (SWAG) function in Arista EOS, Arista relies on Ethernet to group up to 48 campus switches and manage them with just one IP address and a central CLI interface. The classic front panel ports are used for this purpose. Proprietary stack components and protocols are replaced by Ethernet. Switches should appear like a line card in a modular chassis. In contrast to conventional stacking systems, which are often quantitatively limited to 2, 4 or 8 switches and have to be connected with proprietary cables in the ring, SWAG should also offer more flexibility in the campus.

SWAG is designed to support network topologies such as the now ubiquitous leaf-spine design, the more traditional ring or the traditional chain. Smart Software Upgrades (SSU) are designed to simplify stack upgrades by continuing to send LACP control packets even while the switch is restarting, thus keeping link aggregation active. The familiar in-service software updates (ISSU) are also available.

Logical switch management with LSS

The new CloudVision Leaf Spine Stack (LSS) architecture from Arista offers a further management concept for campus networks. Without physically stacking switches, LSS is designed to enable the management of switches via a single logical management plane approach. In addition, LACP link aggregations should also be possible across multiple switches, as has been known from the MC-LAG approach in the data center for many years.

The stack members can also be located in different racks or locations as long as they are connected via Ethernet. Switches should be able to be organized in a hierarchical model, whereby this can represent real constructions such as two switches, a distribution cabinet, a floor with switches, a building or even an entire campus. The assignment to the real spatial situations (buildings and rooms) should be possible via CloudVision. CloudVision Studios, as we tested in iX 06/2024, are designed to simplify the initial provisioning of LSS.

Flexible approach compared to classic stacking

The two technologies are therefore intended to replace previously proprietary stack components, protocols and topologies with Ethernet. The technologies have only been announced for campus switches. The flexible approach is exciting, as classic stacking technologies with proprietary cables are usually limited to a rack or a room. At the same time, however, proprietary cables also offer high stacking bandwidths, even though these are not usually required on campus. Existing virtual stack connections, such as those offered by Cisco on campus, can only accommodate two switches, which can limit the design.

However, whether administration via just one IP address and CLI with SWAG offers the decisive advantage can be doubted in times of automation tools such as Ansible. The issue of "single point of failure" is also likely to play a role in a management plane, although a failover scenario should be available. However, manual management can certainly simplify it. The approach therefore seems exciting. It remains to be seen whether Arista will be able to conquer the campus after the data centers.

(nie)