The data network of the future will be software-defined, so it stands to reason that the software chosen to run the network will be central to the enterprise.
But while the decision to implement SDN or NFV may be easy, nailing down the specifics of the network operating stack is not. While most developers espouse their adherence to open standards and broad interoperability across distributed, cloud-based environments, the specific features and functions needed by key workloads might not be as extensible as one would hope.
Nevertheless, we have to start somewhere, and the networking industry has no shortage of software operating systems and management solution to put the “software defined” into the software defined network.
Pluribus Networks, for one, recently improved the ability to automate the establishment of mesh networks for key advancements like Big Data and virtual desktop infrastructure. The Open Netvisor Linux for Dell, Ver. 2.3.2, extends Pluribus’s Virtualization-Centric Fabric platform to Dell’s Open Compute Project (OCP)-compatible L2 and L3 switches. The intent is to enable a wide range of advanced network management functions, including single-fabric topology creation, application-level awareness and visibility, and high availability across a fully distributed network architecture.
Meanwhile, Big Switch Networks has updated its Big Monitoring and Big Cloud fabric platforms to provide Facebook- and Google-class networking capabilities at the more modest scale of the typical enterprise. The software is available as a free distribution for lab and development environments, although production versions, including those that support 100 Gbps Ethernet, VMware and Openstack environments, will still cost you. And of course, you’ll need the white box hardware to run it on, a line-up that currently includes devices from Dell and Accton, as well as Facebook’s Wedge switch.
Open source communities are also busy upgrading their software stacks as SDN and NFV deployments hit the mainstream. The Open Networking Foundation recently released the second edition of its Atrium stack, incorporating OpenDaylight’s OpenFlow-based controller framework and new capabilities to support Leaf-Spine fabrics. The system utilizes Quagga’s open version of the Border Gateway Protocol (BGP) to enable OpenFlow control plane management for advanced Internet routing, as well as the OpenDaylight Device Identification and Driver Module (DIMM) to provide seamless control across various OpenFlow hardware implementations. In addition, the new version incorporates an experimental version of the Interior Gateway Protocol (IGP), designed to improve the scalability and stability of extended virtual networks.
At the same time, however, legacy network vendors are instituting a steady stream of improvements to their networking software, such as Juniper’s recent tie-up with NEC to incorporate key NFV functions like the vMX routing platform and the vSRX firewall into NEC’s OpenStack cloud platform and NetCracker orchestration suite. The goal is to create a broader NFV package for service providers and enterprises that will provide a smoother upgrade path to distributed, virtual topologies and enable easier management and configuration flexibility once it is in place. Juniper has also partnered up with Affirmed Networks to create a more streamlined virtual mobile architecture, all of which should come together at some point under an integrated network management stack.
The problem with network management and operations going forward is that as data environments become increasingly distributed across cloud-based infrastructure, the need to maintain a cohesive operating environment is hampered by the sheer diversity of hardware and software systems that data will encounter. Even fully open, broadly federated networks will be limited to environments that share their basic formats – although solutions like OpenFlow and OpenStack are likely to be quite substantial.
Perhaps it is wishful thinking to expect a unified, global data ecosystem, but it nonetheless points out the fact that the enterprise needs to take care when establishing the ground rules for their extended virtual networks, unless they are OK with seeing software-defined data silos on the cloud.
