Page 65

EETE OCT 2015

struct a virtual infrastructure with a certain network topology, indicating the way that virtual nodes are interconnected with virtual links. Performance parameters (for example, latency) and resource requirements (such as network bandwidth, compute CPU/memory) are specified in the virtual nodes and links. Generally, virtual resources (nodes and links) are obtained by partitioning or aggregating physical resources. Therefore, a programmable hardware infrastructure is essential to support the composition of virtual infrastructures with high granularity and scalability. In the city environment, the devices deployed in the urban infrastructure are heterogeneous, including wireless/mobile, wired, optical networks, data centers/cloud and functional appliances. In order to enable seamless service provisioning, it’s mandatory to support converged virtual infrastructures enhanced with virtual network functions across the multitechnology, multidomain city infrastructure, so that each tenant can get its own slice of the city infrastructure. However, currently these technology domains are controlled and managed in silos. The NetOS with SDN capabilities at BIO provides a logically centralized control platform that can break through the management silos and bridge the different technology segments. The operating system is able to abstract the heterogeneous city devices, hide their complex technical details and expose the infrastructure in a uniform way. The vision for the white box Open network devices, or network white boxes, use unbranded, generic, modular and programmable hardware platforms. This type of equipment can load customized operating systems and enable on-demand redefining of network functions without the restrictions of vendor-locked devices. Network processors were the initial route to hardware programmability of the underlying network, leveraging the ease of defining functions through software APIs. Network processors are well-known hardware platforms that provide generic programmable features similar to general-purpose CPUs (with extended hardware resources), and can be programmed to perform various networking functions. The main advantage of processor-based architectures is rapid implementation of networking functions using high-level languages such as C, which is highly desirable for rapid prototyping. Network processors, however, are not optimized for parallel operations, which are essential for building high-performance data plane technologies supporting high-data-rate transport. Field-programmable gate arrays (FPGAs) are high-performance and generic processing platforms utilizing programmability from transistor-level to IP-based function level. This makes them highly desirable platforms for designing and prototyping network technologies that must demonstrate high degrees of flexibility and programmability. We are using Xilinx FPGAs that have evolved into systemon chip (SoC) devices in multiple points within the BIO infrastructure: in active nodes - see figure 2 - as optoelectronic white boxes, emulation facilities, wireless LTE-A experimental equipment and IoT platforms. BIO uses programmable and customizable network white boxes that consist of programmable electrical (FPGA) and optical (switching, processing, etc.) parts. These boxes—which enable high-capacity data processing and transport, function programmability and virtualization—are deeply controllable through SDN interfaces. Figure 3 demonstrates the FPGA-based platform, which can host multiple functions in a programmable way, and is interfaced to a programmable photonic part. FPGAs offer several advantages, including hardware repurposing Fig. 3: Bristol Is Open’s network white box is built around Xilinx FPGAs. through function re-programmability, easier upgradability and shorter design-to-deploy cycles than those of applicationspecific standard products (ASSPs). The photonic part of the network white boxes uses an optical backplane on which a number of photonic function blocks are plugged into optical functions such as amplification, multicasting, wavelength/spectrum selection, signal add/drop, etc. Critically, the input and output links are decoupled from any of the functions that the node can offer, unlocking flexibility, efficiency and scalability, and minimizing disruptive deployment cycles with on-service hitless repurposing. Zynq SoC-based Emulation Platform To broaden the capabilities of BIO facilities in experimenting with larger and more-realistic scenarios, we have deployed a network emulator facility within BIO. This platform enables network emulation as well as resource virtualization and virtualinfrastructure composition techniques for advanced network, cloud and computational research. The emulation platform also utilizes local and remote laboratory-based facilities and distributed research infrastructures (networks and computing). Figure 4 demonstrates the multilayer, multiplatform emulation facilities at the core of the Bristol Is Open infrastructure. The emulation facility offers a number of functions instrumental for enhanced network studies in conjunction with the BIO city network and other remote interconnected laboratories: 1. Node and link emulation: This platform can emulate network elements such as routers and switches from the wired and wireless domains, along with the interconnecting links with various physical attributes. 2. Protocol emulation: Whether centralized or distributed, network nodes rely on the protocols to communicate. The emulation facility with precise modeling of the network technologies allows the user/researcher to try out communication protocols and study their behavior on scale. 3. Traffic emulation: Depending on the emulation scenario (wireless networks, data center networks, etc.), traffic patterns with arbitrary intervals and operating from a few megabits to multiple terabits per second can be generated and applied to the target emulated or physical network. 4. Topology emulations: Any topological formation of the desired nodes and links is possible using the BIO emulation facility. This gives the user a chance to fully examine various aspects of the desired technology on the realistic network topologies before deployment and installation. Unlike any other existing facilities that offer computer hostbased emulation environments, BIO uniquely includes programmable hardware (FPGAs, network processors) as well as dynamic and flexible connectivity to multitechnology testbeds www.electronics-eetimes.com Electronic Engineering Times Europe October 2015 45


EETE OCT 2015
To see the actual publication please follow the link above