Issue link: https://cp.revolio.com/i/141720
Figure 2: The services-oriented networking architecture requires a new layer of abstraction and greater autonomy in each network element (NE). End-customer Applications / Services Metadata Resource Database Infrastructure Service Orchestration Abstractions Physical or virtual resource reservation Abstractions Abstractions Abstractions Abstractions NE Forwarding method per service slice NE Traffic Engineering per service slice needs NE State Distribution per service slice NE Management per service slice With the data deluge continuing unabated, workloads will need to become more dynamic as a means to scale data centers more cost-effectively. The two most popular ways for IT managers to handle workloads in a more dynamic fashion are server virtualization and cloud-based services, with latter also utilizing the former. And this steady transition from static to dynamic workloads is already having adverse impacts on data center networks. Consider the impact of just one increasingly common practice: migrating a virtual machine. To perform a VM migration, the hypervisor first reserves a VM on the destination server, then moves the VM to its new destination and finally tears down the original VM. But hypervisors are incapable of generating the requisite Address Resolution Protocol (ARP) broadcasts quickly enough, especially in large-scale virtualized environments. The network can become so congested from the control overhead occurring during a VM migration that the ARP messages fail to get through in a timely manner. Because of the significant impact on network behavior caused by such routine configuration changes constantly affecting connections and routing tables, data center networks will to need a more scalable architecture. Services-Oriented Networking Creating a services overlay to support dynamic workloads requires changes in the data, control and management planes. The 22 | THE DATA CENTER JOURNAL data plane must be able to support dynamic resource allocations and become more flow-centric. This approach requires greater application-level awareness, and to preserve performance as the data center network scales, some means is needed to accelerate packet flows. The services overlay also requires the control and management planes to somehow centralize end-to-end routing/forwarding, which in turn requires more-advanced control, potentially using management-plane proxies. Layers of application programming interfaces (APIs) are needed to centralize and simultaneously simplify various control and management functions. And these functions must all operate independently of the network topology and hierarchy. Service-oriented networking (SON) is an emerging next-generation network architecture that satisfies these needs by enabling each network element (NE) to perform data, control and management functions dynamically and more autonomously on the basis of a set of services. As Figure 2 shows, forwarding, traffic engineering, state distribution and management must all be performed on every service slice by every network element involved end to end. Note the layer of abstractions needed to hide the physical network topology and underlying protocols. In the SON architecture shown in Figure 2, infrastructure service orchestration plays a key role in supporting the dynamic workloads. Effective orchestration requires a dynamic feedback mechanism from the autonomous network element functions to the various service applications; this scenario is depicted in Figure 3. The dynamic feedback mechanism shown in Figure 3 will require the development of some new standards, which are being pursued as part of the SDN effort. SDN is expected to play a prominent role in services-oriented networking for its ability to enable centralized control and management, in part by exposing intelligent abstractions from network substrate. The real advantage of SON with SDN is the ability to minimize or even eliminate the previous hierarchy, and instead use ISO layer-agnostic network elements that are homogeneous and capable of supporting richer services for specific flows dynamically. Figure 3 also shows the infrastructure requirements in the services-oriented networking architecture, and the implications for the next generation of "smart silicon" needed to satisfy these requirements while maintaining adequate performance as the infrastructure scales. Given the ISO layer agnosticism and need for scalable performance, configurability and programmability play crucial roles. In addition, virtualization (i.e., slicing of various silicon-level resources) plays an important role in enabling the SON vision. Within a network element, for example, the queuing, security, deep packet inspection, switching and other such functions must be sliced per service instance. Next-generation communication processors, such as the LSI Axxia product www.datacenterjournal.com