It seems that the farther along we get toward the software defined network, and by extension the software defined data center, the more we confront the issue of connectivity.
This is somewhat surprising, because it has been a steady assumption in IT circles that once SDN moves network architectures to the virtual layer, issues like connectivity will take care of themselves. With no hardware to worry about, applications will be able to forge their own pathways to and through cyberspace and the cloud, often more efficiently than their human overseers ever could.
But it turns out things aren't quite so simple, at least when it comes to building the functionality that supports such dynamic networking. As I pointed out a few weeks ago, connectivity from the data center to the cloud will be a crucial component of the software defined data center, but this is by no means the only way in which connectivity will be implemented. Even a dedicated, high-speed link to a cloud provider is only one facet of what is likely to become a broadly distributed data environment.
In all likelihood, the data environment of the future will be even more distributed than it is now, covering multiple sites either in the cloud or within a wide-area virtual private network. If that is the case, organizations will need some fairly sophisticated software to keep the bits flowing smoothly. Fortunately, a number of start-ups are already addressing this issue, including CPLAN Networks, which recently released the Dynamic Virtual Networks Interconnect (DVNi), which utilizes advanced traffic engineering software to build Layer 2 and Layer 3 VPNs over MPLS networks. The system is OpenStack-compatible and employs label-switch traffic optimization and modeling, as well as automatic discovery and acquisition for all major switch platforms, enabling network environments to be established quickly and easily. It also supports VRFs, MP-BGP, PE-CE and other transactional interfaces to ensure end-to-end service configuration integrity.
In order to keep tabs on all this distributed data, network managers will have to shift their focus from underlying infrastructure to the application programming interface (API), according to MTM Technologies’ Bill Kleyman. At the moment, there seem to be four major connection points between clouds and data center infrastructure that require API compatibility: SaaS, PaaS, IaaS and cross-platform services. But the number of APIs that service these points is growing, with CloudStack and OpenStack contending with solutions from Google, Nimbus, VMware and others. And both Amazon and Eucalyptus are touting the concept of cloud agnosticism in their latest releases through tools like auto-scaling and elastic load balancing.
Indeed, if enterprise networks are to keep up with the dynamism of SDN and the cloud, some long-standing approaches to management and design need to be rethought, says Plexxi’s Mike Bushong. One of the most fundamental is the Shortest Path First algorithm that has guided the industry for more than 50 years. In the future, pathing decisions will need to be based more on resource load, not the number of interconnects between points. By pushing load onto resources that have the greatest amount of available capacity using techniques like Equal Cost Multi Pathing (ECMP), data won’t necessarily follow the shortest route through the network, but it should be the least congested.
The point of all this is that connectivity in the software-defined age will not be based solely on wires or bandwidth or throughput. These will still be important, but full connectivity will also have to accommodate things like traffic management, data and application interfacing, policy management and governance.
As I mentioned above, building connectivity within and between software-based architectures is the hard part. Once the foundations for advanced connectivity are in place, however, the enterprise should find that maintaining a robust networking environment in software is a lot less time-consuming than in hardware.
Photo courtesy of Shutterstock.