TRENDING NOW - GORDON BREBNER
Trending now: Software Defined Networking and Network Functions Virtualization
Software Defined Networking (SDN) and Network Functions Virtualization (NFV) are two trends sweeping the communications industry, both offering the promise of reduced capital and operational expenditure, writes Gordon Brebner, Distinguished Engineer at Xilinx.
Software defined; hardware designed The Field Programmable Gate Array (FPGA) is a cost-effective and power-efficient technology that offers programmable hardware. That is, it can deliver the performance of hardware, but with the flexibility of software. The large, and increasing, sizes of programmable logic arrays (for example, the largest Xilinx UltraScale™ devices now have over 4,000,000 programmable logic cells) mean that very complex functions can now be implemented. The diagram shows the Xilinx SDNet™ methodology for implementing a customized programmable and high- performance (e.g., 100 Gbit/sec line rate) packet processing data plane. The required functions are described in a high-level packet processing specification, which does not require the inclusion of any hardware implementation detail. The overall benefit of such a methodology is that it bridges the gap between the worlds of software and hardware. This stems both from a software- oriented style of design, and from software management of the resulting hardware data plane during operation. In short, the Xilinx programmable hardware is providing a generic, standard, substrate, in the spirit of both SDN and NFV, and the SDNet methodology is then providing ease of use. Protocol Independent Forwarding for SDN The standard OpenFlow control interface was designed with existing switch chips in mind. So, essentially, the nature of
complex services can be delivered using service chaining, where multipleVNFs are used in sequence. For some current NFV use cases, such as for home environments, software implementations may be However, for other use cases, such as mobile core networks or content delivery networks, higher performance and/or more deterministic latency may be required. The solution is the use of hardware acceleration for time-critical functions. In traditional settings, fixed- function accelerators acting alongside processors have been used. However, this approach is counter to the idea of NFV, with its emphasis on virtualization using generic substrates. Instead, programmable hardware offers – as for SDN – a solution that combines performance with programmability. Furthermore, programmable hardware enables service chaining by allowing the dynamic connection of hardware VNFs together in series as required over time. Conclusion Programmable hardware, specifically FPGA technology, together with a software- centric methodology, has the potential to fill the gap between processor performance and required networking performance, while also providing the flexibility needed for future SDN data planes and future NFV virtual network functions and their chaining. sufficient to deliver the required performance.
the packet data plane was pre-determined and, as a consequence, expanding the capability of OpenFlow to control additional network technologies has led to continual modification of the OpenFlow specification. As SDN becomes more widely adopted, for example in optical networking, there has been pressure to evolve OpenFlow to extend the technologies it targets, the protocols it supports, the packet headers it recognizes, and the way in which packets are processed. The result is a new direction: Protocol Independent Forwarding (PIF). The essence of PIF is that the underlying data plane model should be based upon programmable switches, rather than traditional fixed- function switches. There is then an additional operational stage, where a switch is configured or programmed. Once this has been done, the switch is controlled in the conventional OpenFlow style. With programmable hardware, the switch programming mechanism is twofold: configuration can define the architecture of a switch, and custom firmware can then add further features that vary over time. Hardware acceleration for NFV The basis of NFV is the Virtualized Network Function (VNF), originally perceived as a software implementation of a network function that can be deployed on a Network Functions Virtualization Infrastructure. Furthermore,
GORDON BREBNER
T he initial thrust of SDN involved the control of L2/L3 packet forwarding, but more recently has attracted attention in the optical networking community with application to L0/L1 circuit switching for optical transport. The basic idea is to separate control plane functions from data plane functions, so that the former can reside in conventional software running on conventional servers. The latter reside in hardware, due to the performance requirements. NFV is a related, newer, area, where different net working functions are hosted as virtual functions on a standard server, instead of each being provided by separate pieces of networking equipment. As in the case of SDN, NFV is now attracting attention in the optical networking community, for example, for Operations, Administration and Management (OAM) functions. In short, SDN and NFV are about network programmability and network optimization respectively, the overall goal being a more elastic and agile network. In both cases though, with scale-out from the original experimental settings, the reliance on software for providing the programmability and flexibility has resulted in both performance and latency limitations. This focuses attention back on the underlying hardware, and how it can cooperate with software to provide a complete solution.
30
ISSUE 5 | Q3 2015
Made with FlippingBook - Online magazine maker