Cloud Hosting

As 5G, IoT, and other applications look to the edge to satisfy increasing demands, communication service providers are seeking solutions that allow them to operate at the network edge with both high bandwidth and low latency, while maintaining a small physical footprint (in terms of space and power consumption) and reducing TCO.

Software-based solutions are a good start, but they simply cannot provide the required performance with CPU-based hardware. The hardware itself must provide acceleration by way of offloading the data path from the CPU. Lets examine three trends that will become more prominent in 2020 in which hardware-based acceleration will enable edge networking.

#1: Telcos will seek flexibility at the network edge by turning to FPGA-based disaggregated solutions for networking and security

Most data centers have committed to disaggregating their software and hardware, in which they have moved to software-based functions running on top of CPUs inside standard x86 servers. They use open stacks based on Linux software to communicate between virtual machines for application and service chaining while selecting the software application of choice. This gives them flexibility to choose any vendor and ensures the appliance will be futureproof.

Network operators would like to apply a similar disaggregation to network equipment, in order to keep their systems agile enough to keep up with new standards and requirements. However, simply separating the hardware and software is not enough.

Even if telecom companies attain full agility when it comes to software, that only means they can adjust their control plane configurability. As long as they continue to rely on ASIC-based switch silicon, they will be locked into the data plane functionality available on their silicon ASIC (which often lacks many of todays most advanced features), not to mention being locked into a specific hardware vendor.

At the network edge, where protocols and requirements are still evolving, operators must have the agility to adapt to new market demands without having the need to use new hardware when data plane functionality changes. Switch ASICs, though, certainly cannot deliver new functionality and lack the futureproofing that telecoms seek.

FPGAs are the platform that is missing to enable true hardware disaggregation, with complete flexibility in both the control and data planes. FPGA-based networking equipment will let operators achieve true network function virtualization. They provide the performance of ASIC-based solutions, along with a flexible and programmable platform to assure that operators are free to add or change functionality as needed down the road.

FPGAs present a higher CAPEX investment (for now), but they save a lot on operational expenses. Moreover, the need to replace hardware in the field to support new functionality incurs costs significantly greater than the hardware itself. FPGAs will also capture more of the traditional ASIC market as their price drops due to advanced silicon node production.

Brian Klaff,MarketingDirector,Ethernity

In 2018, Microsoft Azure moved from ASICs to FPGAs in their cloud servers in order to keep up with the pace of ever-changing software development. Similarly, in 2020 the trend will be for operators to use FPGAs for switch/routers at the edge of their networks, as they seek to meet the demands of 5G and IoT.

#2: Operators will begin to realize that the performance they are currently receiving from software-based User Plane Functionality (UPF) is nowhere near good enough for full 5G rollouts. They will therefore begin turning to hardware-based acceleration

As 5G gains traction, service operators will be pressed to meet its high performance standards, while keeping their systems flexible enough to adapt to future advances. One of the key pieces to achieving 5G benchmarks is accelerating the user plane function (UPF), which serves as the data plane of 5G networks. UPF is currently handled in software, but this presents several issues, especially as operators consider further improving performance by moving their UPF toward the network edge.

For starters, CPUs are simply not optimized for networking functions, which limits the performance ceiling, regardless of the software employed. Furthermore, many networking and security functions are CPU-intensive, which means several CPU cores must be fully engaged to produce 5G data transfer. This takes up valuable space and power resources at the edge and keeps those cores from being used for the control and application functions for which they are intended.

An ideal UPF deployment would combine the flexibility of software-based virtualization with the performance of well-designed ASIC silicon. The solution is to accelerate the UPF by offloading the data plane to programmable hardware, ideally using FPGAs. FPGA SmartNICs are optimized for networking at the edge, with very low power consumption and a small footprint. Their reprogrammability means that they are flexible enough to handle changes and advances in 5G as the protocols and standards evolve.

In 2020, more operators will turn to FPGA-based hardware to accelerate their user plane functionality at the edge, producing the required performance improvements to achieve 5G at its full potential, including increased bandwidth efficiencies, lower latency, and service-enhancing capabilities such as network slicing.

#3: Telecom operators and enterprises will turn to FPGAs to accelerate SD-WAN deployments as legacy systems struggle to meet new throughput demands

SD-WAN (software-defined wide area network) technology has been instrumental in allowing global businesses to connect and communicate between their headquarters and branch locations. Instead of investing in expensive infrastructure in the form of a private WAN, SD-WAN virtualization has allowed them to operate a WAN over the internet with VPN technology, greatly reducing their cost.

However, there are limits to what legacy SD-WAN systems can handle. As enterprises seek to connect hub-and-spoke installations such as data center to data center, cloud aggregation to endpoints, telecom central office to end users, and business campus aggregation to local enterprise devices, companies are relying more on high-speed communication and need bandwidth over 1Gbps. Under these conditions, legacy systems falter. Businesses looking to meet ever-increasing demand have two choices:

1. Invest in an advanced SD-WAN system that can handle higher throughput; or

2. Keep their existing server/processor system, while offloading the data plane to an FPGA-based SmartNIC to accelerate the SD-WAN solution

Although investing in a new SD-WAN system has its merits, it comes with a large upfront cost and no flexibility in terms of future-proofing the solution. The alternative allows companies to keep their existing infrastructure while instantly and transparently gaining these other benefits of FPGA-based SD-WAN acceleration:

Higher throughput

Reprogrammability to adapt to evolving requirements

Usability with existing uCPE (universal customer premises equipment)

Extremely low latency

Low power consumption

Highly deterministic performance

Support for IPSec security protocol

In 2020, businesses will increasingly adapt their existing infrastructure to meet growing bandwidth requirements by using FPGA SmartNICs to accelerate their SD-WAN solutions.

Read the original here:
Enabling the Network Edge With Hardware-Based Acceleration - The Fast Mode