Optical networking used to be easy. It wasn’t long ago that most applications could predict how much static bandwidth they would need for the coming year, over-provision that amount just to be safe, and then take a few weeks to deploy the resulting network.  While other aspects of the data center were experimenting with innovative new approaches, the data network was a reliable, if uninteresting, part of the corporate infrastructure. That was before workloads became mobile and dynamic, before data became big and social, and before the economics of cloud computing completely disrupted the industry across practically all market verticals. It soon became apparent that if servers and storage were reaping huge benefits from abstractions and virtualization, then data networks also needed to change with the times.

Faced with this seismic upheaval, the market has responded in many different ways. Some companies tried to deny that change was happening, or claimed that they had been doing business this way all along. This argument has become harder to sustain, given the emergence and rapid growth of OpenStack, CloudStack, OpenDaylight, and many others. Others have recognized the business value in dynamic networking, and embraced the wave of change which is still sweeping the industry. Fundamental change creates opportunities if the risk associated with it is analyzed and understood. This is where New York State Center for Cloud Computing and Analytics at Marist College plays an important role in verifying the benefit of programmable connectivity for future cloud data centers. 

The SDN/NFV Innovation Lab at Marist has demonstrated real world use cases for dynamic cloud networking on their test bed, a 125 km metropolitan area network linking three enterprise data centers with the ADVA FSP 3000 wavelength division multiplexing platform. This test bed was among the first in the industry to implement dynamic orchestration of optical wavelengths, a function which is only recently being formalized by open industry standards. In an early demonstration, open source software was used to monitor CPU utilization in a cloud data center, and trigger dynamic bandwidth allocation to enable virtual machines to move from a congested data center across the MAN to another location with available capacity. All the switches within the source and target data centers, as well as the optical WDM between data centers, were provisioned from a common software defined network (SDN) controller. This demonstration showed that a process which would normally take days or weeks using conventional systems could now be completed in minutes or less using SDN. Potential use cases include disaster recovery and business continuity, and this approach compliments the use of virtual machines and NFV functions in the data network.

In a subsequent demonstration at last year’s SDN World Congress, it was shown that storage volumes could also benefit from dynamic wavelength allocation, triggered by traffic monitoring embedded in the WDM equipment, rather than inside the data center. This approach solves a long standing problem related to synchronous storage replication. While many organizations are moving to a hybrid cloud model because it’s much more cost effective than adding storage resources to their own data center, the remotely attached cloud storage still doesn’t behave exactly as if it were locally attached. Some applications require large amounts of storage access in a short time. Measurements at several production data centers suggest that these bandwidth spikes can last 15-30 minutes and exceed normal traffic levels by 3-5 times or more. Failing to keep up with these bandwidth spikes can cause storage controllers to time out and applications to fail. It’s not practical or economical to address this issue using static, over-provisioned bandwidth, but this problem can be addressed if we add bandwidth on demand from a pool of wavelengths whenever a traffic burst occurs.

Bandwidth slicing, the ability to subdivide a physical network into multiple virtual network segments with their own network controllers, is a very attractive application of dynamic wavelength orchestration. This allows a cloud service provider to allocate network resources for each tenant, while giving tenants the flexibility to manage their own slice of bandwidth. Each tenant can select any industry standard controller, and we’ve demonstrated this approach in both KVM and VMWare hypervisor environments. Multi-level control plane designs such as this are the first step towards enabling cloud exchange services, the next step in cloud evolution which acts like a telecommunication exchange to handle data traffic.

The ability to provision optical networks from mobile platforms such as a tablet or smart phone is a further feature which can be easily and seamlessly provided on top of a programmable network. The importance of mobile apps for the enterprise and cloud has received increased attention recently, due to announcements such as the Apple – IBM alliance. This approach is expected to further reduce both capital and operating expenses, and it’s one example of how cloud economics have impacted the industry, creating lower barriers to entry offered by cloud architectures, and the conversion of fixed costs to variable costs. Optical networking also stands to benefit from the same elastic capacity approach used in cloud computing. 

All of this has made networking much more central to an organization’s business strategy, and provided unprecedented opportunities to innovate, transform, and grow optical data center networks.  A new generation of optical networks emerges, which offers not only higher bandwidth but many other benefits derived from programmability (including automated service activation, improved reliability, security, and energy efficiency). And through the use of SDN management abstractions, these new features can be realized with user-friendly, graphical interfaces that make the network easier to use than ever before, and a lot more productive than ever before. Demonstrations as performed by Marist College are an important step from feasibility to commercial application.