The Evolution Of The Ethernet Switch update from February 2014

Buying an Ethernet switch is no longer a simple task. While there are several reasons for this, perhaps the greatest one is that of innovation. Today, buyers must learn a variety of technical nuances that set switches apart from one another, match those capabilities to their organization's needs, and then move ahead to a purchase.

If you think faster speeds are the extent of the new Ethernet technology, you're focusing on one detail of a much larger picture. Yes, greater speeds are an important part of what's new with Ethernet switching; 10GbE and even 40GbE are found in a wide variety of platforms. But beyond the speeds themselves, new forwarding technologies that enable all links in a layer 2 domain to be utilized have become commonplace. Beyond that, the networking industry is looking ahead to software-defined networking (SDN), where traffic can be forwarded through a datacenter in ways we haven't even thought of yet.

In this article, I'll examine how Ethernet switching is evolving. It's a whole new landscape out there.

Speeds & Feeds

Raw carrying capacity is perhaps the most easily observable change Ethernet has seen over the last several years. 10GbE is now ubiquitous, finding its way into a variety of applications. 1GbE access layer switches often feature 2 or 4 10GbE uplinks for connectivity out of the campus closet or top of rack. Many top-of-rack switch models feature a high density of 10GbE ports, with 40GbE uplinks. And there also are switches supporting the 10GBase-T standard coming to market, as datacenters consider running 10GbE over copper instead of fiber.

The great deal of additional speed facing hosts (along with hardware that can fill the bigger pipes) has necessitated faster switch internals as well. High-end data centers switches are often non-blocking, meaning that they can forward bi-directionally at full-line rate across all ports. Even if a 10GbE switch is not non-blocking, reasonable oversubscription ratios between host-facing and uplink ports make for a switch that can move an astonishing amount of data.

Switch vendors have also made strides in the area of port-to-port latency, with the highest performers measuring the amount of time to switch a frame through the chassis in hundreds of nanoseconds.

Fancy Forwarding

In the realm of forwarding traffic, the industry has made great strides with technologies that allow all layer 2 links in a switch to forward traffic, with no links placed into a blocking state by spanning tree. The most common approach is multi-chassis link aggregation (MLAG), which allows two or more physical switches to present a group of parallel links as a single aggregated link to a neighbor. Using industry standard LACP to present the link bundles, MLAG allows for fully redundant switching infrastructures that can interface with any other LACP-capable device.

Leaping beyond MLAG, we find layer 2 fabric technologies TRILLand Shortest Path Bridging (SPB). These protocols allow a sort of routing of layer 2 Ethernet frames across a meshed fabric, where the shortest path between hosts is computed and then followed. While similar in concept and even execution, SPB and TRILL are incompatible technologies. Furthermore, those vendors basing their fabrics on TRILL are incompatible with each other, with their proprietary variants. SPB has followed a tighter regime of interoperability testing among vendors, with broad agreement as to implementation of the SPB standard.

[Find out why vendor licensing requirements make it hard to switch to TRILL in "TRILL's Hidden Cost."]

Ethernet switches have also become an interesting convergence point for storage traffic. As Fibre channel SANs are still prevalent in many data centers, Ethernet switches are often capable of forwarding FC frames inside of Ethernet, a technology known as FCoE. This allows datacenter operators to reduce the amount of cabling that must be fed into a host, sending both data and storage traffic through a converged network adapter. Ethernet switches capable of FCoE are also capable of bridging to a FC SAN.

Software-Defined Networking

The growing SDN movement has certainly impacted Ethernet switches. This is especially true in the datacenter space, where bringing IT services online quickly through automation has reached into the network tier. Switching vendors are making their hardware more accessible, usually by exposing the configurable elements of their gear through APIs. While APIs by themselves are not an automation solution, they are a key building block that more and more vendors are supporting.

Of course, OpenFlow is still making its disruptive presence felt, although with somewhat less fervor than perhaps the industry experienced a year ago. Even so, innovative networking applications that rely on OpenFlow continue to be demonstrated as a part of open-source projects such as OpenDaylight, or in vendor proprietary SDN ecosystems.

Despite a slowing adoption and silicon implementation challenges across the industry as a whole, OpenFlow continues to matter as an emerging way to program switch forwarding tables. Indeed, vendors such as NEC and BigSwitch have bet big on OpenFlow, producing switches, controllers, and applications that rely on OpenFlow to push application forwarding policy down into the physical hardware.

Switching is a huge area of innovation for the networking industry, with vendors continuing to find ways to genuinely differentiate themselves from each other. Not every switch is right for every potential customer, so it's essential that you understand the changing landscape before making a purchase.

This article is the beginning of a deeper exploration of the Ethernet switch market I will publish on my blog at ethancbanks.com as a multi-part series entitled "The Ethernet Switching Landscape." I will also present this deeper look at Interop Las Vegas in two sessions: "The Ethernet Switching Landscape Part 1: Matching Technology With Problems" and "The Ethernet Switching Landscape Part 2"I hope you'll follow along! I welcome your comments, questions, and discussion.