Tthere are parts of the NPU market that will do well and others that will not. The edge of the network, with its burgeoning application growth, provides a fertile cradle for the NPU model while the core network is so harsh an environment that even the best managed companies face long odds of generating superior investment returns. Let’s examine why that is.
The Application Agnostic Network
Carriers and their customers are moving away from TDM based services, switching, and transport. The technology that will replace this infrastructure is IP and Carrier Ethernet. No debate is needed on these two points. This transition has the added benefit of creating networks that are application agnostic – unlike TDM, which was purpose built for voice and re-purposed for everything else.
This emergence of the agnostic network is creating a new era of network applications. Consumers experienced this first as the web era and later with P2P file sharing, YouTube, online gaming, etc. Enterprises see it in the form of datacenter consolidation, Enterprise VoIP, and an explosion in Software as a Service (SaaS) applications like Salesforce.com. And applications emerge every day.
This variety of network applications is possible because an Ethernet Wide Area Network (WAN) connection can be re-purposed for any application whatsoever, provided the WAN connection has a well documented quality of service profile. Use the WAN for Email and VoIP by day, offsite backup by night. The faster the connection becomes, the more applications that become possible. Make the connection 10x faster and suddenly multiple corporate data centers can be combined in a single geographic location, accessed over the WAN instead of the LAN. It is impossible to predict the ultimate gains in productivity and cost savings a radical increase in network speeds to the Enterprise will create, other than to say they will be large.
Heterogenous Edge & Homogenous Core
Faster connections at the edge of the network will result in an explosion of innovation as new applications are invented; security, caching, network acceleration, and storage virtualization are just a few areas. But these new applications (and the countless applications that have yet to be invented) are possible only if these heterogenous applications can assume the availability of WAN connection with consistent quality and behavior.
Carriers, equipment makers, and chip vendors recognize this requirement and are spearheading multiple standardization efforts to create what is called a Universal Network Interface (UNI). We’re not taking sides on which proposal is the best, but what is clear is that carriers and the supply chain that feeds them know their survival depends on making this happen.
The UNI requirement will result in carriers building networks that aren’t necessarily stupid (as David Isenberg famously proposed), but consistent. Equipment from node to node, from supplier to supplier, must be homogenous in function to make this happen. Superior execution of a well understood and agreed list of requirements is valued. Flexibility, extra features and special enhancements have little value, or worse will result in equipment that doesn’t conform to the norms of the transport network. This would yield an inconsistent experience to the user at the edge of the network, and defeat the purpose of the UNI.
A consistent UNI also allows customers to evaluate carriers with a common service denominator. Carriers will compete on their ability to deliver this UNI based on cost and network breadth. The successful carriers will be the ones that can deliver a consistent UNI in the most locations for the lowest cost.
The key takeaway: A homogenous network yields the consistent interface (UNI) that heterogeneous applications at the edge of the network require. Building a successful homogenous network requires that carriers stamp out complexity by constructing and managing their networks in a uniform way as inexpensively as possible.
It turns out that removing network complexity is also the easiest way to drive cost out of the transport network. Carriers accomplish this by pushing the complex application infrastructure to the edge of the network, where it resides at (and is maintained by) the customer, or is concentrated at a few central nodes of the carrier. As new services are invented and deployed, only the edges of the network must change, reducing the cost and complexity of introducing new services. The core transport network that connects the two edges isn’t necessarily a dumb pipe – but it is a consistent one – and isn’t subject to feature modification.
This trend towards placing the complex application infrastructure at the edge of the network, and enforcing consistency and simplicity in the core is shaping the component and equipment industry. Companies that can adapt to the asymmetrical distribution of complexity between the homogenous core and the heterogeneous edge will succeed. Those that cannot will fail.
The Value of the NPU
NPU’s add feature flexibility to network equipment when compared with their alternatives, usually fixed function silicon (commonly referred to as application specific standard products ASSPs). But this flexibility comes with additional costs, and must be justifiable.
As a result, NPU’s are well suited for use at the edge of the network, specifically for providing the flexibility needed by ever evolving applications. The opposite is true in the core, where functionality is static, and the flexibility NPUs provide is a cost liability.
At the edge of the network, where these heterogeneous applications exist, NPUs face nearly limitless opportunity. The situation there is much like the early stages of the computing evolution. However, NPUs are a liability within the transport network itself, where homogenous behavior is required, flexibility and intelligence are not needed, and simplicity and cost dominate.
No other component is impacted by this carrier trend of pushing complexity to the edge.
In the next installments of this series we’ll look at the make (ASIC) vs. code (NPU) vs. buy (ASSP) decision process equipment makers face, some of the interesting competitive dynamics of the NPU market, and where existing public and private companies fit into the edge/core market structure.
Companies discussed in this series: Alacritech, AMCC, Bay Microsystems, Broadcom, Cavium, CSwitch, EZ-Chip, Fulcrum, Marvel, Mindspped, MIPS, Raza, Tarari, Xelerated.