Building Fabrics on SDN Architecture
Defining the network in software is not the only challenge facing the enterprise this year. For many key applications, those software architectures will have to resemble mesh fabrics as well.
Granted, this is much easier to do in software than in hardware, but it seems likely that, initially at least, software defined fabrics will have to work with legacy physical and virtual fabrics in order to provide a high value proposition.
According to Technavio’s latest report on network fabrics, the overall market is set to grow by about 7 percent per year for the rest of the decade. This is not as robust as other segments of the data stack, but it does point up the fact that organizations are looking to derive maximum flexibility within their overall network environments. Data center fabrics are particularly popular among cloud-facing organizations because fabrics allow them to offer a range of advanced services that could otherwise only be deployed within the data center at a great cost of time and expense.
Pluribus Networks is one of the first SDN vendors to incorporate a fabric-style architecture into its platform. The new Virtualization-Centric Fabric (VCF) option on the Netvisor operating system provides a high degree of elastic connectivity for virtual workloads and their related distributed services. This allows Netvisor to support a virtual fabric running on open switch hardware capable of tracking the entire network’s endpoints and flow patterns. The platform features port and virtual network segmentation for combining physical and virtual resources under a common framework while automatically maintaining fabric connectivity as it scales up and out. It also supports common programming interfaces like REST and Java to present a single fabric front-end for applications.
Fabric architectures are also emerging on the data center interconnect as organizations seek to leverage advanced public and private cloud constructs. Arista Networks recently added new fabric capabilities to its EOS and CloudVision solutions aimed at providing long-haul connectivity that is more flexible and less costly than current deployments. The upgrades include the Spine Transit 100G DWDM line card and the Spine Interconnect VXLAN fabric extension system that allows disparate fabrics to be connected under a single Layer 2 architecture. The system offers the possibility of converting remote backup and other secondary facilities into co-equal components of a distributed storage and/or compute cluster.
Hyperscale operations like Facebook are also steeped in both software definition and fabric networks, and now the garden variety enterprise has a chance to emulate those architectures. Facebook recently released the design of its ToR Wedge switch through the Open Compute Project, right down to the list of materials needed to build it. At the same time, Facebook has offered up its FBOSS operating switch, which can be run as an Open Network Linux image on software platforms like Big Switch to create a full programmable networking solution. When deployed as part of the company’s Six Pack aggregation platform, users will be able to create a custom network fabric consisting of up to 12 Wedge switches in a single chassis. At the moment, the Wedge switch tops out at 40G, although a 100G version is expected soon.
Fabric topologies are most effective in highly complex environments such as analytics in which data from multiple disparate resources need to be compiled quickly and efficiently. As such, the enterprise has no real need to implement an end-to-end fabric across the entire data stack.
But as Big Data and the Internet of Things adds to the enterprise workload in the coming years, it’s good to know that software defined, or at least SDN-supported, fabrics are moving into the channel as well.