In the case of the conventional, modern networking industry, standard bodies such as IETF would define standards for the networking and storage industry. Vendors, in turn, would implement their solutions accordingly, but this would often result in compatibility issues (and undermining the whole point of having standards in the first place).

In order to restore compatibility between black box implementations, vendors previously spent weeks engineering. Then, equipment and software designers would congregate in large gatherings called “Plug Fests” to test and validate actual hardware and software for interoperability. But this practice -- the process of getting new standards approved and proving interoperability for every new application -- would be seen as absurd and impractical in the server computing world. So why is it still a common practice in networking?

This is partly because networking standards have been in place now for three decades. And, these networking standards were previously driven by telecom use cases, where the downside risks from failed systems already deployed were just be too catastrophic to ignore.

Yet this process consequently caused significant delays for vendors to go to market. Software needed to be tested, validated, reworked in the firmware, re-tested and then re-validated, often delaying the product release itself.

Today, Linux has become a mainstream operating system for servers, but not so for networking. This is in part due to lock-ins by proprietary networking vendors, and black box implementations of networking applications and services, networking operating systems (OS), and hardware.

In conventional networking, the data plane and control plane are implemented within the firmware of routers and switches. With the two working in harmony, the data plane carrying the network user traffic, this enables data transfer to and from clients, as well as the handling of multiple conversations through several protocols including VXLAN, GRE, and the management of connections with remote peers. The control plane carries signaling traffic, and, in routing, defines what to do with incoming packets via the use of routing tables and implementation of preferential treatment for packets like QoS for differentiated services.

In turn, the whole systems is essentially a locked down, black box implementation.

Software-defined networking (SDN) is changing this. SDN decouples the control plane from the data plane and implements the control plane in software, thereby disaggregating the black box implementation. This means that networking applications, network OSs, hardware can come from different vendors as long as they can “talk” to each other using each others’ prescribed interfaces deployed on branded white boxes or “brite” boxes. White boxes are built with off-the-shelf switching chips known as merchant silicon, and loaded with open SDN software to run many networking applications.

The story gets better. With Linux making inroads as the network OS atop standards-based x86 servers, the method of communication between the network application and network OS is standard Linux protocols, which have always been “open.” Having Linux as the network OS changes the networking landscape, as new applications running on the Linux-enabled switch do not need an IETF standard to talk between each other, just like any application running on Linux on servers within the data center.

Unlocking the control plane with open interfaces opens the doors to radical innovations for the networking space. A new generation of networking applications previously trapped by specialized ASICs (which take years to build and cost millions to design and fabricate) can be re-imagined as a minimal viable product from mint startups, as the adoption of Linux-enabled switches becomes mainstream.

The data plane performs important functions such as enabling data transfer to and from clients, handling multiple conversations through several protocols as discussed above, along with the management of conversations with remote peers. However, the data plane is still constrained somewhat by firmware and ASICs. The reason is because vendors that control the data plane like Broadcom and Intel are keeping the interfaces between the network operating system and the merchant silicon closed or with limited access.

While network application development remains complex, with Linux gaining traction as a network OS, developer tools and libraries for Linux are now broadly available. Regardless of the complexity, there also seems to be no shortage of Linux experts in the global community ready to solve problems and lend their expertise. In contrast to conventional networking, developers for a given platform are typically found in enclaves only amongst a few employers.

Conventional network applications had to be qualified individually for every supported network OS. With Linux, a network application can be written once and can run anywhere. Cycles previously spent on platform qualification (such as interoperability testing) can be redirected to solving age-old problems in networking previously trapped in firmware/ASICs, unlocking a whole new generation of software-based networking tools in automation, security, overlays and fabric managers. While traditional networking do not have a pan-data center awareness, Linux opens the door to a new generation of network virtualization solutions that can harness the power of the fabric and the overlay to give network operators full control of the physical and virtual environment.

SDN solutions achieve the same objective by encapsulating lower level protocols into higher level objects. Networking virtualization solutions implement their own inter-communication within the virtual network.

In use cases where the cloud operator is trying to connect clouds using two disparate systems, or between two geographically disperse data centers, or between a public cloud and a private cloud, one would use a standard like BGP to connect the two cloud infrastructures.

The oligopoly of conventional networking is about to be broken up by vendors in every layer of the networking stack and replaced by more nimble, innovative software vendors. Age-old problems previously hampered by firmware and ASICs can be solved by running Linux-enabled switches. New entrants leveraging open source and commodity components are more likely to pass along savings to end-users to garner market share. Moving from conventional standards held by a few proprietary networking vendors to a disaggregated model is a win-win for end-users and for the industry.