Apr 02, 2013
Once the purview of consumer-facing companies like Google, Amazon and Facebook, access to information—anywhere, anytime, and from any device—is increasingly required from enterprises of all sizes. Widespread adoption of mobile devices, a growing volume of rich content and higher bandwidth rates in both 3G/4G mobile and wired networks have only accelerated this demand. Of course, the precise level of demand varies greatly over time and across users and applications. For computing, virtualization solutions have proven an excellent match for dynamic provisioning requirements within enterprises, delivering improved efficiency and higher utilization.
Conversely, networking infrastructure has lagged in the area of virtualization. Network planning has traditionally meant overprovisioning—i.e. deploying equipment capable of handling the highest possible traffic that could pass through that network node. Yet in today's highly dynamic, cost-conscious and rapidly expanding data center environment, there is a great need to cost-effectively and intelligently utilize networking devices.
The buzz on software-defined networking (SDN) as the solution, with its central coordination aspects and potential to dramatically re-architect the networking landscape has reached an all-time high. However, adoption volumes remain relatively modest. Why is this?
To start, let’s get a sense of the SDN landscape. Broadly defined, it’s an industry-wide effort to separate and centralize control plane elements from distributed data-plane elements in networking equipment. The purpose is to enable flexible and dynamic provisioning of the data-plane to adapt to varying traffic with proper security, virtualization and quality of service (QoS) across multiple tenants and applications in a cost effective manner. A significant number of vendors have rallied towards use of the OpenFlow specification for SDN, which describes functionality in the data-plane (switch) and an interface between a centralized controller and the data-plane. The growing SDN ecosystem now includes switches (Pluribus), virtual switches (VMware), L3-L7 services (Embrane), orchestration products (NetCitadel), routers and access points from a range of vendors.
A parallel but complementary effort, network function virtualization (NFV) initiated by 13 major carriers is driving a shift towards virtual software appliances rather than purpose-built hardware networking appliances that have been popular over the last two decades. NFV enables these virtual SW appliances to run on powerful multi-core, commodity server hardware.
The promise of SDN & NFV concepts is significant, with the following benefits for customers, similar to what has been achieved by compute virtualization:
Lower capital expenditures and operating costs through sharing of equipment across multiple tenants and applications
Flexible and dynamic deployment of networking resources with security and QoS
Ability for users, applications or networking services to scale performance up or down elastically based on demand
Modular and standard software architecture and APIs enabling multi-vendor solutions and a rich ecosystem of third party solutions
Fault and diagnostic capabilities not available in existing networks
Greater feature velocity and innovation
SDN: Major hurdles to overcome
While the vision of SDN is very attractive, it faces several challenges that could slow or hinder widespread adoption:
Networking Industry Incumbents: Today, major vendors provide a complete stack of networking hardware and software. While their products interoperate at the network packet level, the management and provisioning of their devices plus certain services are proprietary, enabling higher margins and customer stickiness for the vendors. The leading market share vendors, for obvious reasons, don’t like a significant industry transition that could negatively impact their share or profits, and are doing their best to cautiously steer it in their favor.
Interoperability with legacy, deployed equipment: To achieve widespread adoption, new networking equipment needs to work well in existing networks. If the full promise of SDN requires forklift upgrades, that is not going to happen! Companies like Nicira (acquired by VMware) recognized this and provided a compelling overlay solution that is compatible with existing infrastructure. However, more work is required to fully exploit all the capabilities in current and upcoming SDN-enabled equipment, such as network virtualization, security and QoS features in various data-plane nodes.
Incompatible, different standards and the business goal to differentiate hardware: The OpenFlow switching specification has done an excellent job with data-plane standards and southbound interfaces between the centralized controller and the data-plane. However, different vendors are adding proprietary extensions to OpenFlow to differentiate their solutions, which could limit multi-vendor deployments. Additionally OpenFlow itself may evolve and unless all equipment is future-proof upgradeable, different OpenFlow versions will co-exist in the network. Finally, standardization for northbound interfaces, between the centralized controller and the application or management services, is still in its early stages.
Confusion and lack of compelling value for new customers: While network and server administrators are excited about SDN’s vision and are monitoring this trend, their first job is to support existing networks and businesses. Without a compelling near-term value proposition in cost, performance or manageability, they are unlikely to quickly and widely adopt SDN.
Widespread SDN adoption: Learning from other platform successes
Consider platforms that have achieved widespread industry adoption, like the x86 PC/Server or iPhone/Android smartphones or, in the networking space, Ethernet’s win over Token Ring and other proprietary connectivity solutions. Ultimately, each evolved into a stunningly successful platform with well-defined interfaces, a thriving ecosystem, rapid innovation and broad adoption.
However, in each of these cases, the initial product did not emphasize the platform or ecosystem as the primary value proposition. Instead, the focus was on the product’s tremendous benefit. The PC democratized computing by providing low-cost access to businesses and consumers. The iPhone delivered a huge leap in smart-phones through touch interface, larger screen and a complete browser. Its initial version did not even support third party applications. Both classes of devices fit into existing infrastructures without forklift upgrades. With Ethernet, it adapted to the inexpensive and ubiquitous twisted pair wiring and provided compelling advantages in cost and usability.
In sharp contrast are platform- or framework-focused initiatives that flickered but failed to ignite. Remember OpenWave HDML/WML, which shipped in large volumes in handsets but were never widely adopted? Or Taligent’s object-oriented Pink OS for PCs? In the enterprise, DCE and Corba both tried and failed. Considering the hurdles to SDN adoption—industry incumbents, legacy equipment, compatibility challenges, and a tough sell when it comes to near-term value—these examples seem particularly instructive.
The lesson? Companies should develop SDN products that deliver huge near-term value to the customer with ease of deployment, while conforming to the overall SDN frameworks and specifications. Such products have the ability to ship significant volumes, which in turn will attract third party developers. Interestingly, there are opportunities for multiple products and companies to create platforms at different levels, ranging from L1-L3 products (e.g. Pluribus) to virtualization/control layers all the way up to intelligent provisioning of the data-plane elements (e.g. Embrane). Companies that do this first will be major winners, while others will end up supporting them as part of the overall ecosystem build-out.
Rajiv Khemani is an Executive-in-Residence with NEA.