Rethinking the Network

Michael Bushong

Subscribe to Michael Bushong: eMailAlertsEmail Alerts
Get Michael Bushong: homepageHomepage mobileMobile rssRSS facebookFacebook twitterTwitter linkedinLinkedIn


Related Topics: Virtualization Magazine, Intel Virtualization Journal, Intel SOA Journal, Cisco Virtualization Journal, SDN Journal

Blog Feed Post

Silicon: Merchant vs. Outsourced

When people talk about merchant silicon, most of the time they fail to really clearly specify what it is they want.

If you talk to folks in the networking industry about switching hardware, you probably don't get more than a few sentences in before you hear the words merchant silicon, typically followed closely by commodity hardware. There is some confusion in the dialogue about these topics, and I think it stems from a lack of clarity around why this matters and how silicon development actually works.

Let's start with why merchant silicon matters.

When people talk about merchant silicon, most of the time they fail to really clearly specify what it is they want. There are basically two reasons those on the buy side ought to care about merchant silicon: interchangeability and cost. I will focus on the former in this particular post.

si-waferAs I have written about previously, interchangeability is the highest bar for what people want when they talk about Open. That is to say that customers crave the ability to replace one device with a functionally equivalent device from another vendor. The rationale is that two equivalent devices will force both vendors to innovate faster and stay honest on price, while also providing the buyer with choice and flexibility. If the hardware platforms are all custom made, functional equivalence is much more difficult to guarantee, so the thinking is that consolidation on a common chip set will make this interchangeability more likely.

And so ends theory.

In practice, I think this has been historically true and is moving in the direction that it will be only marginally true in the not-too-distant future. For the first few years of merchant silicon development, platform requirements were clear and common. Essentially, merchant silicon vendors needed to provide basic functionality that existed in their custom counterparts. In the switching space, Cisco was the dominant player, so Broadcom had to replicate basic switching functionality that had been ubiquitously deployed by Cisco customers. There is no way that Broadcom (or any silicon vendor) could re-create every piece of functionality in existing custom chip sets, so they prioritized foundational features.

During this time, development is fast and furious. It is relatively more straightforward to re-implement existing capabilities, so the effort and time-to-market are lower. This is partially why early designs are cost-effective. Doing almost anything a second time is simply easier than the first time. And cost is obviously an important aspect of developing a market for merchant silicon.

But what happens when all the low-hanging fruit has been plucked?

The basic set of switching requirements is growing over time, but certainly not at a breakneck pace. You could argue that overlay support is a new addition, but there isn't a ton of additional common functionality required across the entire industry. So once the foundational stock is all in place, who generates demand for new features?

It certainly isn't the merchant silicon vendors. It would be insane to think that they could create new capabilities and generate enough demand across the set of systems suppliers. And being a component vendor, they are one step removed from the end customers, making it that much more difficult to act as a central point of new feature delegation for the industry.

Clearly the drivers of new functionality are the networking gear makers themselves. But once the common features are all retired, do they collectively work from a single list of requirements? Obviously not. Each networking company will be sourcing its own customer demands and designing products to meet those demands. If every vendor had the exact same view of how to solve problems, we would end up with a universe of identical products (at least appeasing the Commoditize All the Things contingent).

So the dynamic we end up with is a set of merchant silicon consumers (the systems companies) driving individual requirements to the merchant silicon vendors.

There are a couple of ways this can play out. It could be that Broadcom listens to its set of customers, identifies those requests that are common, and only builds those. This would mirror what most consumer companies do. They survey their entire customer base and build what they think is most broadly applicable. But the market dynamics in the networking industry are different. With a very small set of network silicon buyers, Broadcom will be particularly sensitive to its largest customers.

Imagine a scenario where Broadcom's largest customer demands Feature X. Broadcom says "We would love to, but that feature is not common across the rest of our buyers, so we simply cannot do it." Intel gets wind of this and immediately engages with a promise to build whatever feature the customer wants. Broadcom then loses its biggest customer - and not for a year, but for an entire generation of products.

This does two things: first, Broadcom loses the immediate business. But more importantly, Broadcom loses the volume, which forces them to raise the price on a per unit basis for everyone else. This upsets one of the major drivers behind merchant silicon: cost. The risk is simply unacceptable. Broadcom cannot say no in this situation, so their only alternative is to build the feature into their next chip rev.

What just happened is subtle but very important. The chip consumers are driving the roadmaps (at least in part). So while there will always be general work applicable to everyone (performance enhancements, scale, SDK, and so on), there will also be an increasing amount of custom functionality designed with a particular network gear maker in mind.

This kind of development is really not merchant silicon so much as it is outsourced silicon.

The meaningful distinction is not whether the resources reside within the networking company or whether they are effectively contract resources reporting into some other corporate entity. The question is one of customization. To what degree are networking gear makers influencing silicon roadmaps for their own products? Because once this happens, the original premise that all solutions are interchangeable is no longer valid.

What's a customer to do?

First, I want to be clear that these dynamics are healthy and expected. And they are certainly not unique to Broadcom, so I don't mean this to call them out in particular. But these dynamics are real. Because the undercurrents of Open and Merchant Silicon are so strong, particularly as SDN, white box switching, NFV, and network virtualization all emerge, customers will need to be aware of the industry dynamics and how they impact long-term product trajectories. There will almost certainly be long-term (likely hidden) impacts to interchangeability (in other words Open) as well as cost.

Customers who are aware will be in a better position to ask intelligent questions and more accurately assess the potential impact to their architectural plans.

The post Silicon: Merchant vs. Outsourced appeared first on Plexxi.

Read the original blog entry...

More Stories By Michael Bushong

The best marketing efforts leverage deep technology understanding with a highly-approachable means of communicating. Plexxi's Vice President of Marketing Michael Bushong has acquired these skills having spent 12 years at Juniper Networks where he led product management, product strategy and product marketing organizations for Juniper's flagship operating system, Junos. Michael spent the last several years at Juniper leading their SDN efforts across both service provider and enterprise markets. Prior to Juniper, Michael spent time at database supplier Sybase, and ASIC design tool companies Synopsis and Magma Design Automation. Michael's undergraduate work at the University of California Berkeley in advanced fluid mechanics and heat transfer lend new meaning to the marketing phrase "This isn't rocket science."