Posted: May 3, 2011
First things first. Why start a blog?
Simple. This is the most exciting time in networking in the last decade. As I describe briefly in the about page, there are so many technical and economic vectors straining the traditional network model that change is inevitable. Here are a few examples:
- The cloud cost model is built around commodity supply chains. Yet networking gear is expensive. Damn expensive. Whether or not it’s relatively more or less costly than other parts of a data center, it’s still damn expensive. I’ve seen quotes of multiple hundreds of millions of dollars from incumbent vendors for new build-outs. On the other hand, a black-art is emerging in which cheaper gear is being used to build highly functional data centers. The price difference with a traditional vendor are staggering. Take for example Nick’s presentation at this year’s Interop in which he includes a back-of-the-envelope calculation approximating a savings of $400 million across 10 data centers. As we continue to slip into a technical service economy, cost reduction will continue to grow in importance and affect not only the supply side of networking, but how they’re designed as well.
- The rise of merchant silicon as a vaible alternative to special purpose ASICs cannot be overstated. I’ve had opportunity to work all of the silicon vendors, Broadcom, Marvelle, and Fulcrum, and their silicon is awesome. It’s clear to me (and many others) that the position proprietary silicon has is attenuating, and merchant silicon will likely be switching the bulk of tomorrow’s packets. Not only are companies like Arista building awesome products that are spanking the traditional vendors, new and very aggressive startups are popping up that will offer gear at true OEM prices. Watch for it.
- Compute virtualization is giving rise to a slew of new networking products and standards. These includes those purpose built for virtualization (Nexus 1000v, VMWare VDS, Citrix Distributed vSwitch Controller, VNTAG, EVB, etc.), and many created for other purposes that are being pressed into service (Trill, LISP, and many existing NAC and network management products). What I find particular exciting is that the formation of this ecosystem is not technology looking for a problem, but rather very real problem looking for the correct solution. That’s a great recipe for genuine innovation.
- Software Defined Networking is picking up steam. I’ll cover SDN in much more detail in the future. But roughly, it’s an umbrella term for architecting networks such that the control plane can be implemented as a distributed system as opposed to the traditional protocol-driven model.The implications of SDN are many. Most importantly (in my mind) is that this allows developers outside of the vendors to build bad-ass control software. I’d rather much rather join an open source software effort than argue with a bunch of “lifers” in the IETF, and then wait a decade in the hopes that a vendor will implement a compatible solution (which is unlikely).
The sum of these trends is real, macro-level pressure guiding networks to be come proper systems. Finally. That means, horizontal integration, true virtualized abstractions, real APIs, software lead innovation, open platforms, etc.
Make no mistake, this isn’t a transition, it’s a transformation. Or perhaps even, a revolution!