FPGA-based network acceleration for 5G network slicing
by Blogs & Opinions
5G network slicing
Network slicing is not a new idea but is taking on greater importance as 5G services roll out around the world. Network slicing provides a single physical infrastructure to support multiple "logical" or "virtual" networks. In addition, network slicing enables a network to virtually look like separate, dedicated networks – each with its own capabilities from the perspective of bandwidth, latency, resiliency, privacy, security and isolation. Network operators can partition a network for different users or "tenants" to multiplex bandwidth and services over one infrastructure.
Network slicing is implemented for various reasons – the most basic is to provide a dedicated set of resources that are guaranteed to a customer or application. These resources may be as simple as isolation from others to more sophisticated divisions for guaranteed throughput per customer, guaranteed uptime, or low latency transport.
While important in many virtual network overlays, network slicing is particularly interesting in the case of 5G because of the varying applications that will be using this infrastructure for transport and the requirements inherent in these varied services over 5G.
5G technology and services will enable use cases never before achievable over a mobile network.
Home Internet access, IoT and mobile broadband all have very different requirements from a network perspective. For simple Internet access or mobile broadband, customers will demand guaranteed uptime and a guaranteed amount of throughput. For IoT, on the other hand, network providers will need network slices requiring lower throughput, but extremely low latency for sensor-to-sensor communication at the network edge to support millions of endpoints, sensor locations and short-lived flows. Other application use cases may be transactional in nature, where a tenant wants to "dial up" or "dial down" their network resources based on time of day, product launches, events or innumerable other dynamic, business-related issues. Effective network slicing makes all of these cases possible.
Network adaptability – a system's ability to change itself dynamically and efficiently based on elastic circumstances – goes hand-in-hand with network slicing. Network slicing enables adaptability by carving resources into individual networks, each controlled individually. Each of these "virtual" networks can be modified at a moment's notice to change their path through physical networking elements, and change redundancy characteristics and/or per-tenant performance requirements. The adaptability offered through virtual network slices is key to making network slicing an effective technique to support varied service level agreements, application requirements and security profiles.
Requirements for network management
However, the question of how to manage a network at this level of granularity must be considered. Network operators simply cannot manage the massive scale and breadth of their infrastructure on a per-device basis. Network slices, by definition, are end-to-end constructs that span the mobile and physical portions of the network connecting wireless node to wireless node, to edge infrastructure and to the broader cloud or Internet.
This level of connectivity requires programmatic control of all these resources in a seamless fashion regardless of the physical location of the network equipment. Administrators require a holistic view into the entire network from a logically single location to "see everything" and change the entire behavior of the network from a single touch point. There are countless techniques and technologies existing and in development that enable the dynamic nature required to manage, monitor and secure a virtually defined, network-sliced environment.
While admirable goals, the vision of operational simplicity and dramatically lower total cost of ownership while maintaining performance is still a dream on the horizon. Operators have realized that to scale applications in 5G virtualized networks to meet performance goals they will need a new data processing paradigm to realize the requirements that 5G applications place on network infrastructure.
Application processing, networking QoS guarantees, latency, security, encryption/decryption, flow processing and network monitoring are all different from a workload perspective. Different workloads require different processing strategies. To support the networking requirements of network slicing in virtualized environments we must look at solutions from a different perspective.
Data plane acceleration
To scale virtualized networking functions (VNFs) to meet performance goals and provide network slicing transparently operators need to implement data plane acceleration based on FPGA-based SmartNICs. This technique offloads the x86 processors that are hosting the varied applications to support the breadth of services promised.
Functionality required by 5G network slicing is network-centric by definition. These networking and security workloads are poor prospects for general-purpose architectures. Varied workloads should be handled on the right processor type at the right location. This proven architectural solution tightly couples general-purpose processors and FPGA-based SmartNICs in a workload-specific fashion. The architecture utilizes general purpose compute for application processing and leverages SmartNICs for data plane and security processing. This technique alleviates the data plane-related workloads from the CPUs to fully utilize all resources and support monetizable 5G applications.
Software-configurable, FPGA-based SmartNICs specialize in functions like virtual switching, flow classification, filtering, intelligent load balancing, QoS and encryption/decryption, which can all be performed in the SmartNIC and offloaded from the processor housing the 5G applications. With technologies like VirtIO these functions can be transparent to the application and provide a common management and orchestration layer to the network fabric for network slicing. Coupling FPGA-based SmartNICs with general-purpose x86 or ARM processors allows operators to realize the vision that 5G and network slicing promise.
Overcoming the challenges of 5G virtualized deployments that require network slicing means using reconfigurable computing platforms based on FPGA-based SmartNICs and standard servers. These accelerated servers will be capable of offloading and accelerating compute-intensive workloads – either in an inline or look-aside model. This will appropriately distribute workloads between the general-purpose processors and software-reconfigurable, FPGA-based SmartNICs optimized for virtualized environments.
About the Author
Daniel Proch is Vice President of Product Management at Napatech. His role includes technology strategy, network architecture, and hardware and software planning. He has 20 years of industry experience spanning product management, CTO's office, strategic planning and engineering in previous roles at Netronome, Ericsson, Marconi and FORE Systems. Daniel has engineering and telecommunications degrees from Carnegie Mellon University and the University of Pittsburgh.