Technology vendors love to grab terms that are hot and then overuse them to the point where no one really understands what it means any more. I understand the desire to catch a market trend and have the “rising tide” lift the vendors along with a number of others. But the overuse of terms tends to confuse buyers while they are trying to figure out what’s what.
This is one reason why Gartner’s Hype Cycle has the phases it does. While I think some of the terms are a little silly, the fact is that the first upslope creates vendor overhype and then technology goes into a lull while users do their own research.
If you’ve been around the network industry for a while, you probably remember the days when the term “stacking” became such a term. There’s some debate as to who invented stacking.
I remember both SynOptics or Digital introducing products in the early to mid 1990s that would improve the resiliency and manageability of fixed form factor switches through stacking. The term originally was used to describe a feature that enabled management of an IP cluster of a “stack” of fixed form factor switches. Network managers could mix and match technology, create redundant links, aggregate bandwidth and do a number of other tasks to the stack. However, very quickly, any vendor that could daisy chain a bunch of products together claimed to have stacking technology even though it wasn’t close to the original definition of it.
+ Also on Network World: IDC: SD-WAN market to hit $6B by 2020 +
Most recently, the term “software defined” has caused a great deal of confusion, particularly with software-defined WANs (SD-WANs). Last month, I wrote a post discussing how everything from broadband WAN to WAN optimization to bandwidth management now falls under the term of SD-WAN.
Recently, I’ve noticed the term “fabric” going through the same challenge while vendors try to promote their network fabrics as a core part of their SDN strategy. This begs the question: what is an Ethernet fabric?
Fabrics were built to eliminate many of the limitations of virtual LANs (VLANs), Spanning Tree, traditional routing and even virtual private LAN service (VPLS). Don’t get me wrong, these technologies were great innovations in their time, but the world has changed and they can’t support many of the requirements in a modern data center. For example, vMotion from VMware created the need to support East-West traffic patterns, which most legacy networks cannot do.
I believe the first protocol used to build an Ethernet fabric was something called TRILL (Transparent Interconnection of Lots of Links), which would enable a standardized way of building a data center network where traffic could go from point to point with a single virtual hop. In practicality, it may pass through lots of network devices, but it doesn’t need to do lookups and other network actions at each step, so it looks like a single hop.
How fabric should work
A key requirement of a fabric is that it should work across vendors, allowing for traffic to move seamlessly across different environments. Any kind of translation that has to be done at the edge of a network would break the uniformity of the fabric.
There are a couple of vendors today that support TRILL, but they either have proprietary control or data planes, making them non-interoperable with “standards-based” TRILL. It may be possible to get these proprietary versions to work together through some sort of TRILL gateway, but as a I point out above, this is no longer a single fabric. The solutions do address the limitations of legacy networks and offer fabric-like capabilities in a single network, but its use would be limited to that one network.
A few years ago, both the IEEE and IETF agreed on an Ethernet fabric standard known as shortest path bridging (SPB – IEEE 802.1aq and RFC 6329). A handful of vendors, including Huawei, Alcatel Lucent Enterprise and HP, have adopted this protocol to be part of their fabric. Avaya, through the acquisition of Nortel, also has an SPB solution and has been aggressively promoting interoperability of SBP-based solutions.
As I mentioned before, Ethernet fabrics were developed to overcome many of the limitations of legacy networks, particularly in next-generation data centers to support the explosion of East-West traffic. Now with SDNs, it seems a number of vendors are stretching the definition of a fabric, which has the potential to cause confusion and slow down both SDN and fabric deployments.
Another important point to consider is the growth of hyper-converged platforms. These platforms require a scalable and robust Ethernet fabric to scale, so when vendors position solutions that require VLANs and routing tables, it seems the term “fabric” is being used liberally—similar to how “stacking” was used several decades ago.
There’s a Latin phrase, caveat emptor, that means “let the buyer beware,” and that certainly has applied to network infrastructure. If you’re looking at investing in a network fabric, make sure you purchase the right technology and not just terminology. More and more solutions claim to be a fabric, making it confusing.
A good starting point is my post that looks at the Network of 2020. Fabric solutions today should include most of these features, and the vendors should be able to at least give you a roadmap as to when they will be able to provide a full solution.