From Sand to Software – My VantagePoint

Eric Broockman Chief Technology Officer Published 26 Nov 2019

As the CTO of Extreme Networks, and given that part of my career has been in developing chips for communications equipment, I get asked from time to time, “Why doesn’t Extreme do their own networking chips like Cisco or HPE?”  My Vantage Point here isn’t specific to Extreme. It is really an answer that most customers should think about, no matter who their network vendor is. Said differently, if a customer buys a product that is built on proprietary silicon rather than merchant silicon, then over the course of the ownership of the product, they will come to recognize that they have made a strategic mistake; one that is going to cost them in terms of productivity and total cost of ownership to the detriment to their enterprise. (Think proprietary CPUs vs. X86 for instance. How’d that work out?)

So how is that, you might ask. If you look at one of the aforementioned companies, their most recent product family is built as a derivative cost-reduction from a chip family that has been around for many years. Yes, there are a few new cute additions, but essentially the end switching product is built on a cost-reduced version of yesterday’s news. Does it work? Sure. It does a fine job of delivering the features of yesteryear, and its future trajectory will, therefore, be incremental enhancements to yesteryear as well, because that is the heritage – the DNA so to speak – of this technology. Don’t get me wrong, it does have some solid features, but let’s dig deeper.

Let’s compare this to merchant silicon. First, a merchant semiconductor supplier has an inherent advantage over any product company’s internal chip team with the exception of unicorns such as Apple. In particular, the nature of a merchant semi supplier is that they interact with all potential users of their product. They listen to the use cases and trends from top customers and consolidate and synthesize all of this input. Additionally, the top merchant semi suppliers, such as Broadcom and Intel, routinely meet directly with the end customers who buy the products with their chips in them as well. This gives them direct input from the top customers that drives the forefront of their products, whether they are hyperscale data center companies, top financial institutions, government entities, etc. Overall this results in better input to the merchant semi suppliers and logically, better products result as well.

In addition, in this era of networking, the true market leaders setting the trends in networking are not actually the top few network equipment suppliers. Rather, the trendsetters are and have been for some years now, the hyperscale cloud and top social media giants. Think of the hyperscalers as being at the top of the food chain. This means that today’s hyperscale capability will be tomorrow’s financial service product foundation and the next day’s service provider enterprise solution. Interestingly, the silicon used by ALL of the top hyperscalers and financial institutions, and essentially all of the top service providers, is Broadcom merchant silicon. Why? Because the merchant suppliers listen better, they execute better, they push the technology envelope, and they deliver new chips, from top to bottom, far more frequently than proprietary vendors. For instance, Broadcom delivered 3.2T, 6.4T, 12.8T faster, with better features and more reliably than the networking companies and their own internal efforts.

Get the latest stories sent straight to your inbox!

Therefore, if an enterprise wants to have its own infrastructure benefit from all of the advances in networking being driven by the top dogs in the industry, then they should purchase products built from chipsets that are smaller and enhanced versions of chips used by the hyperscalers. For instance, Broadcom’s family of Trident-3 chips, targeted at the enterprise and used across Extreme’s own products and roadmap leverage these capabilities. This includes advanced telemetry features, programmable capabilities, sophisticated encap/decap features, and more. Additionally, these new chips preserve feature capabilities that extend back over a decade – thus protecting our customers’ installed base.

Now let’s circle back and look at the consequences of using networking products that rely upon internally-developed proprietary “lock-in” chips. Here is what happens. A new product family comes out. It is based upon a new proprietary chipset. The network OS includes some cool marketing features that only work with the internally-developed chipsets. The customer gets convinced (conned?) into using one of the features. Then, the system company falls behind leading competitors that are using merchant silicon – so they buy a chip startup company to try and catch up. They build some new chips. There is a 90% overlap with the earlier generation of chips and the cool new “catch up to merchant features” – but these new chips lack full compatibility with the gee-wiz features that the customer got conned into from the earlier generation. Now if the customer wants to add some new gear with the cool new “catch up” stuff into their network, they have all sorts of challenges – because the new gear isn’t fully backward compatible with the old gear. Sound familiar? This in turn artificially results in a refresh cycle that has to come sooner than it would otherwise be required, not to mention a higher cost of ownership. A real time example of this phenomenon happened about a week ago with the launch of some new cloud-native products by a major industry player. During a chat Q&A, this question came up: “Is the CLI the same as LastGen or is it different?” The condensed answer was “there are many similarities to LastGen, there are also differences . . .”  You get the idea. 

By contrast, merchant silicon providers generally behave like Intel with regards to providing backward compatibility of CPU features. This enables the network equipment manufacturer to deliver new enhanced versions of its network OS which are indeed truly backward-compatible with the installed base of gear. This means lower cost of ownership and less artificial network refresh cycles. So, if you like lock-in, if you like higher refresh costs, if you like inferior chipset capabilities, if you like less choice, and if you want a network OS that can’t deliver the same benefits enjoyed by the world’s most demanding networking – then buy from the proprietary lock-in vendors. Otherwise, do what the smartest networking buyers in the world do; what the hyperscalers do, what the world’s largest banks do – buy products built upon merchant silicon. It will save you money over time and delivers superior results in your network. Arista, Extreme, Google, Microsoft Azure, and AWS are all great examples of companies that build the hearts of their networks upon merchant silicon.

Related Stories