Redefining Networks to Keep Up with the Cloud
If your house is too small for your growing family, you remodel, and if your network is not capable of dealing with the demands of cloud computing services, big data analytics or technology advances, you segregate and automate it with Software Defined Networking (SDN).
Both residential and business networks are deploying more traffic, from an increase in social networking to the increased use of broadband video. The so-called millennial generation (i.e. anyone born after 1985) doesn’t think twice about uploading an image to a community or followers. Other generations are not far behind however, consuming cloud services and uploading data. Last year, users of Snapchat uploaded approximately 760 million images each day, close to WhatsApp users’ 700 million each day (over 8,100 per second) and far above Facebook’s daily total of 350 million and Instagram’s 70 million images a day. That inclination to upload on social media, combined with video feeds and “visual networking,” where social media annotations are layered over broadband video for interactive content, is taking its toll on data centers. The increased data and storage demands, and subsequent bandwidth increases have placed pressures on data centers to improve efficiency, while environmental and real estate costs are driving the need to reduce space, cost and power consumption.
“By separating the network control plane and the forwarding plane, the former can control several devices and the configuration can be tailored to the needs of a particular enterprise.”
To meet these demands, Software Defined Networking (SDN) is the architecture chosen by system developers. By separating the network control plane and the forwarding plane, the former can control several devices and the configuration can be tailored to the needs of a particular enterprise, whether healthcare, retail, telecommunications or IT, for example. The Open Networking Foundation (ONF) defines SDN thus: “The architecture decouples the network control and forward functions, enabling the network control to become directly programmable and the underlying infrastructure to be abstracted for application and network services.”
The Foundation is developing open standards, such as the OpenFlow Standard, which is the only vendor-neutral standard communications interface that sits between the control and infrastructure, or forwarding layer, of the SDN. It is also working on the OpenFlow Configuration and Management Protocol Standard and working groups are developing interoperable SDN concepts, frameworks, architecture and standards, in collaboration with SDN experts and suppliers.
SDN can manage the networking infrastructure, automating data traffic management and managing big data, which requires considerable levels of parallel processing—which needs additional capacity as it is distributed to multiple sites and analyzed for decision making within an organization. Creating adaptive, intelligent networks can allow managers to control complex networks for analytics, while manging resources and data traffic. Crucially, it also allows new services, such as Cloud computing, and adding devices onto a network (known as Bring Your Own Device, or BYOD). The ability to modify the physical networking environment into software also reduces the overall capital expense (CAPEX) and operating expense (OPEX) as organizations modify existing networks. All of these contributors account for the growth of the SDN market worldwide.
The current increase in demand, and that which is still to come, means that the SDN market will see a Compound Annual Growth Rate (CAGR) of 47 percent between 2016 to 2022 (Allied Market Research – Global Software Defined Market Opportunities and Forecast) to reach a value of $132.9 billion by 2022.
The Cloud is a significant factor in the SDN market, says Ron diGiuseppe, Senior Strategic Marketing Manager, Solutions Group, Synopsys. “SDNs are introduced in response to Cloud data centers’ large, virtualized workloads,” he says. They are designed to meet the needs of extra capacity required by operators, for example, on Cyber Monday, the Internet’s busiest online shopping day, or when they expect increased capacity demand, for example around events like the World Cup or Super Bowl. Extra bandwidth needs to be added to the network to add capacity and scale on demand. Users also demand fast response times, with no buffering, so that these interactive networking experiences do not suffer delays. The virtualized workload increases the level of traffic and the flow of horizontal server-to-server traffic, says diGiuseppe. “Virtualization means that traffic increases and the workloads get distributed,” he explains. “Instead of north-to-south traffic flow, the virtualized data workload means server-to-server traffic from east to west.”
In the SDN, the server chips need increased bandwidth capacity, and Synopsys DesignWare IP provides building blocks to integrate this into the System on Chip (SoC), together with high-performance peripherals, says diGiuseppe. The company’s DesignWare IP includes DDR3/DDR4 controller and PHY IP, for memory choices, PCI Express 4.0/3.1 Controller and PHY to enable virtualization for servers, at low power operation, 10G and 1G Ethernet controller and PHY as well as 40G Ethernet controller and PHY IP, as the means to configure the SDN, accommodating low latency by offloading the server Central Processor Unit (CPU). It also offers a security protocol, such as AES (Advanced Encryption Standard) IP for infrastructure security.
In a typical SDN there is a 64-bit host processor on a server System on Chip (SoC) with Double Data Rate (DDR) memory, Peripheral Component Interconnect Express (PCI Express, or PCIe) bus, Ethernet communications, but also hardware accelerators, for example, a security accelerator for infrastructure security. The accelerator has to communicate with the host processor, creating a trend, identified by diGiuseppe, that of the heterogeneous accelerator, which allows optimization engines to implement different functions in hardware.
The heterogeneous interconnect must synchronize with the processor, requiring the interconnect to be cache coherent to connect both hardware and software to accelerate co-processing. Part of this movement is the CCIX Consortium, which includes foundry TSMC, Synopsys, Xilinx, ARM, and IBM amongst its members. The consortium aims to produce a standard for a chip-to-chip interconnect that allows two or more devices to share data in a cache coherent manner. Importantly, the standard will allow processors based on different instruction set architectures to seamlessly share data and extend cache coherency to accelerators, interconnect and I/O. In this way, heterogeneous components from multiple vendors can be combined to meet a system’s requirements. There are no CCIX products available yet, but the consortium expects members to introduce them beginning this year.
Integrated Development Environment
One of the members of CCIX Consortium, Xilinx, has developed an Integrated Development Environment for SDN. Software Designed Specification Environment for Networking (SDNet) creates packet processing systems, based on user-defined specifications to the company’s All Programmable Field Programmable Gate Arrays (FPGAs) and SoCs, and creates SDNs.
High-level networking specifications are processed by a set of integrated development tools which allow the user to describe the desired packet processing functions.
“The main benefit of SDNet is that it increases productivity by implementing design at the higher level, without going into the details of hardware design,” explains Awanish Verma, Senior Manager – SDN, Network Function Virtualization (NFV) and SDNet – Product and Technical Marketing, Xilinx. It automates the verification and validation process, he continues, and reduces the typical design hardware cycle drastically. Verma estimates that the typical length of time for writing description language, verification and validation can be reduced to one to two weeks, compared with one month to write the hardware and a month to verify a module.
Equally important, is the fact that SDNet is device-independent. “You don’t need an Xilinx device to implement SDNet,” adds Verma. “[SDNet] sits on top of Xilinx design implementation tools. . . generating a new list, to bring a hardware list to the software implementation.”
Gilles Garcia, Director Communications Business, Xilinx, adds that the environment brings hardware and software teams together. This is partly practical, as more software students are graduating from universities. Garcia believes the networking space will be driven by software engineering, as there are around 85 percent software graduates compared with 15 percent hardware students graduating from universities today. SDNet anticipates this ‘bias’ “providing powerful abstraction layers that do the hard work for hardware, helping software engineers to run the hardware, without having to learn the hardware language.” It is the software team that drives the deployment and the concept of the packet processing for virtualization, he says. If a packet behavior is incorrect and needs to be modified, the software team can change parameters without having to learn the hardware language.
Caroline Hayes has been a journalist covering the electronics sector for more than 20 years. She has worked on several European titles, reporting on a variety of industries, including communications, broadcast and automotive.