The Day Amazon Ended the Hardware-Defined Network
Amazon is one of those companies that does what they say. They recently completed their examination of SDN technologies and made their deployment decision moving away from vertically integrated, expensive systems in favor of standards-based systems and software.
While vendors and the academia have been pushing SDN aggressively, network operators have been cautious in transforming their networks toward less vertically integrated solutions. This is understandable, networking hardware and the software systems that control them, are a complex and proprietary discipline that only a few, specially trained individuals can truly understand and deploy.
That used to be true in the data center, but not any longer. Today, businesses can buy server hardware from any vendor, virtualize it with any hypervisor and then run any application on it with full control and visibility as well as upgrade various components at any time. The benefit is maximum control and flexibility across a low-cost infrastructure.
But when it comes to networking the very same network operators are trapped within vertically integrated nodes from companies desperate to maintain an “economic system” that drives high-cost and complexity into modern communications. The end result, of course, is the limited ability for network operators to get control over the cost and complexity of their networks and the inability to react to customer-demand for elastic resources.
Why is the CapEx so high?
In a vertically integrated solution such as a switch or router, almost all the components are custom made and operated with proprietary software that only talks to other proprietary modules. Essentially, these devices are not built on standard high-volume (SHV) components or common off the shelf (COTS) parts running software that can be replaced with alternatives or use industry standard protocols. Even though networking is moving toward “software-driven” design and concepts – the large vendor incumbents continue to highlight the value of their software on their hardware. The result is an expensive vendor lock-in and high-cost systems.
Why is the OpEx so high?
As the cloud, remote applications, distributed computing, content delivery, and mobile has grown, the technology to run and operate networks has become more complex. From protocols, to hardware, to software suites controlling that equipment, to operational and support systems—the bloat has caused increased operational complexity and cost. Specialized network engineering personnel have to be trained with proprietary knowledge to maintain and communicate with these systems. In addition, service and support contracts are required to keep up with device change and configuration and to manage network failures related to this complexity.
What do users want?
Networking must catch up with the cloud and the data center. Users are demanding the disaggregation of software from hardware, more network elasticity and control, and open methods of interfacing with software and hardware. They want to use merchant silicon and general purpose CPUs to implement switches, routers and all the other network functions. Furthermore, they want to be able to run any software on any hardware and use any network management software to control and orchestrate the network. But most importantly, they want the network to get out of the way and to be able to run applications at their full capacity to fulfill business needs.
So what is the industry doing to react? One example…
Recently, Cisco launched Insieme’s product line and laid out a vision for its own version of SDN: an application centric network that automatically adjusts the network characteristics to meet the demands of the applications – as long as you buy brand new Cisco hardware and build your entire network out of Cisco gear. While Cisco is on the right track by focusing on the performance of the applications and the delivery of services and solutions rather than the technology underneath, not surprisingly it is advocating a tightly integrated model that offers very little OpEx, CapEx or multi-vendor advantages.
A disaggregated model allows each layer (silicon, hardware, software, API, etc.) to evolve and innovate at its own pace.
Need proof? Look at the PC: CPU, motherboard, bios, OS and application… in most cases, each company focuses on building the best product for each layer. The time for the likes of Sun and IBM are over.
Need more proof? Look at the data center, as discussed above.
A closed, vertically integrated model stalls innovation and progress as we have seen time and time again, particularly in the telecommunication provider space.
Operators are looking for a new age that includes increased control and flexibility, reduced dependence on vendor-specific high-cost hardware, and a simpler operational model.
What does the future hold?
Disaggregation of software from networking devices and improved multi-vendor interoperability gives the control and visibility back to the users. This is inevitable. Why are we so confident? Because we have seen this cycle in the past. Network equipment is the last place in the industry where you have to buy the hardware, software and support from the same company. Disaggregation and ultimately, SDN, enables the service differentiation, control, and elasticity that service providers need to compete and to react to a new set of dynamic end-customer requirements.