Network functions virtualization (NFV) and software defined networking (SDN) concepts have laid a strong foundation in preparation for more network transformation. This is a reality for enterprise, cable, mobile, fixed and media networks and their operators. Now CommSPs are extending NFV and SDN to the network edge to place more compute, storage and network capacity closer to end-users and their devices. These edge investments in cloud-based infrastructure are necessary to support 5G commercial deployments and the next-wave of innovation in new services and business models.
Networks must be robust, service-aware and cost-eﬀective to support applications on the path to “cloud-native”, that will operate independently and abstracted from the physical infrastructure. These applications will rely on the availability of secure, scalable and adaptable network architecture. 5G-ready networks require architectures that scale up with the requisite performance, power, footprint and integrations. The network architecture should also scale out to disaggregate data, network compute and functions with cloud and cloud-native concepts. Network automation is the final tenet required to orchestrate the secure movement of data.
Scale Up with Intel Architecture and New Platform Capabilities
Intel continues to expand its portfolio of high performance server CPUs to support a range of performance, power and space requirements across the network. We offer high-performance, low-power system-on-a-chip (SoC) processors for high-density computing in space and power constrained environments, processors for cloud and data center environments and solutions in-between. Built on the Intel® Xeon® Scalable processor architecture, an advanced compute core optimized for a broad portfolio of balanced platforms, these solutions deliver powerful performance and scalability for cloud-optimized and 5G-ready communications networks.
With Intel® Architecture, CommSPs are able to converge applications and services, control plane processing and high-performance packet processing workloads. This convergence offers greater agility and lower total cost of ownership while laying down a stable foundation for NFV and network transformation
High packet throughput and improved end-to-end latency, quality of service (QoS) and traffic management are required to support 5G services. These requirements demand the efficient movement and sharing of data between virtual network functions (VNFs), containers and microservices. This coordination or orchestration of data requires hardware queuing and platform resources, including cache, memory, I/O and accelerators, dedicated to data plane applications in a highly granular fashion. Intel is driving new platform innovations in these key areas in support of 5G infrastructure transformation.
Scale Out with Cloud infrastructure
The industry is asking for a network infrastructure that evolves into disaggregated pools of compute, storage, networks, I/O devices and accelerators, allowing CommSPs to construct network applications on demand. The construction of a stateless, distributed application architecture based on the cloud native concept will require a robust platform infrastructure that can scale with agility and tolerate multi-point failures.
In order to meet the needs of service providers for cloud native solutions, there are three areas of development that Intel is influencing and driving:
- Kubernetes Networking
- Data Plane Engine & Abstracted Interfaces
- Service Meshes
- Kubernetes Networking: Containers have been around and deployed for quite some time. However, CommSP requirements and the need for networking support bring new challenges and complexity. VNFs often require more than the one network interface provided by Kubernetes due to the separation of the management/control and data planes. Multus Container Network Interface (CNI) plugin is one example of a mechanism to identify the network(s) to which interfaces should be provided for, in the Kubernetes pod.
- Data Plane Engine & Abstracted Interfaces: Data plane VNFs require performant network interfaces to achieve data line rates and minimum latency with optimal server resource cost. NIC SR-IOV (single root I/O virtualization) and vhost-user are two CNI plug-ins created to address these needs in the Kubernetes environment. The NIC SR-IOV CNI plug-in enables SR-IOV virtual function (VF) allocation to a Kubernetes pod in both kernel and user space networking modes to accelerate network performance for packets ingressing/egressing the server. The vhost-user CNI plug-in allocates to the Kubernetes pod a network interface with a userspace vSwitch backend (e.g. OVS-DPDK, FIDO/VPP), accelerating the network performance of communications between Kubernetes pods on the same server. These CNI plugins have been demonstrated to provide high performance network interfaces in flexible network configurations for a variety of NFV use cases.
- Service Mesh: The Service Mesh concept is widely deployed in Cloud Native environments. While we support SR-IOV and Multus in Kubernetes, the Network Service Mesh represents a scalable architectural approach. For Layer 7 Services, Istio is a service mesh with significant adoption.
Intel’s data plane efforts include the ability to accelerate Envoy in a way that is transparent to the application. We also support industry efforts, including the Ligato open source project, to create a Network Service Mesh where layer two and three (L2/L3) services can be coordinated in a cloud native way. While still early in the development phase, we believe this kind of architecture promises to solve many of the problems the industry faces in deploying cloud native network functions.
Network Automation: Tie It All Together
Network Automation, including Service Orchestration, is another key tenet of NFV and transformed networks. ONAP, OSM, OpenStack and Kubernetes are open source projects focused on solving the problem in a multi-cloud environment, but we have a lot work ahead. Intel is contributing in two key areas:
- Enhanced Platform Awareness (EPA): Intel has made a concerted effort to expose key platform capabilities to the Management and Orchestration (MANO) stacks to address the TCO challenge faced by the Service Providers. EPA allows the service providers to create, deploy and manage an end-to-end service with a full knowledge of platform capabilities. This awareness allows them to place the workloads and applications where they can deliver optimal performance and improve total cost ownership.
- Over the last 2 years, Intel has standardized how telemetry and statistics from various platform components, such as CPU, chipsets, NICs, storage and memory, are exposed to the orchestration systems and applications. Intel selected Collectd as the tool of choice to expose this valuable platform telemetry and create plugins into various open source projects. Intel envisages ultimately is that will be able to apply Machine Learning to this telemetry data in order to recover from fault scenarios in the network infrastructure without human intervention and auto-configure the network based on dynamic resource utilization.
Let’s Collaborate and Get Networks Ready for the 5G Future
In conclusion, transforming networks to truly deliver next generation of innovative services is necessary for the 5G future. Intel has been a pioneer and industry leader contributing to many industry standards, open source and ecosystem development efforts.
Please join me at the Open Networking Summit in Amsterdam to continue collaboration discussions on how we can partner and drive the network transformation to deliver 5G to the market.
Follow me on Twitter @rgadiyar or visit www.intel,com/network for more information.