5G is slated to usher in an unprecedented increase in throughput, reduction in latency, and support for high device density. Besides this, and more importantly, from vertical industries’ point of view, it supports differentiated and deterministic services, unbounded by technology and carrier boundaries, along with a goal towards ubiquitous presence. One of the most critical technology enablers of these 5G offerings are the two networking technologies Network Function Virtualization ( NFV ) and Software-defined Networking ( SDN). While these two technologies evolved independently, by working in conjunction with each other in different ways, they provide the necessary agility and flexibility to make rendering of 5G services technically and economically possible for modern day Communication Service Providers (CSPs).
NFV leverages the base virtual infrastructure built on top of actual physical infrastructure (data centers) consisting of compute, network and storage resources to rapidly deploy or modify a network solution through deployment of Virtual Network Functions (VNFs). SDN uses the concept of separation of control and data plane to enable differentiated end-to-end processing of transmitted data packets. While these leverage the concept of data centers, there are some unique requirements specific to the raison d'être of a communication network viz. transmission of information across geographically separated locations. This is in contrast to traditional data centers’ hosting different applications whose services are consumed by different users and the involved networking is not the focus.
Some key design considerations one needs to take into account when designing data centers for CSPs are:
Focus on networking
The CSP’s focus is on efficiently providing information transmission across Service Access Points (SAPs) to consumers and with minimal distortion. These SAPs are distributed across geographical locations, which require the CSP’s capability and presence to be distributed. So, unlike a traditional data center, the aim is not to centralize services into a central location, but to create a distributed data center solution which allows CSPs to render the services efficiently across their geo scope. Also, as the focus is on connectivity, the solution is more about providing appropriate switching capabilities, rather than traditional compute and storage capabilities.
Meeting varying communication requirements
5G is expected to unleash a plethora of services which would require deployment of a wide variety of networking applications {which could be VNFs, Physical Network Functions (PNFs) and Hybrid Network Functions (HNFs)} at specific geo locations. This would require appropriate design to ensure efficient rendering of services (e.g. multi-purpose yet specific resources at Edge Data Centers). Also, the designed data centers must consider the existing network solution as well as the transformation road map.
Proximity to target locations
5G services requiring ultra-low latencies need to move processing of information as near as possible to the target locations. Mobile Edge Computing (MEC) based solutions, which enable real-time applications, will need to run on appropriate compute resources. The data centers at the edges need to be designed to support envisaged demand for MEC-based 5G services. The DCs must allow for execution of third-party application specific services.
Support 5G slicing
5G slicing enables differentiated communication services aligned with specific requirements of various industry segments. Dynamic slicing of available network requires changing deployments of VNFs at different data centers depending on anticipated demand. The 5G distributed data center solution needs to be architected so that VNFs’ deployment location can be modified. The distributed data center must balance availability of physical resources on demand and efficient usage of the same.
Need for accelerated processing
While a majority of the functionality required for effective rendering of communication services can be met using generic COTS hardware, there are some specific needs that can be met only through accelerators, which are specialized processing units. These are expensive and need to be judiciously placed and connected along with the generic hardware.
Affinity and anti-affinity requirements
To be efficient, certain packet switching functions rendered through VNFs will need to be as closely placed as possible to reduce latencies. 5G is expected to provide low latency services and services with SLAs on latency. This will require that the physical transmission time be minimized as it drives affinity requirements between VNFs. Guarantees on availability of services is key to the uptake of 5G services by industry verticals. This will require provisioning for redundancy and the distributed data centers need to be designed so that anti-affinity requirements between VNFs and communication paths can be enforced. The DCs must be designed to provide the possibility of alternate paths between SAPs and a thorough failure mode analysis must be carried out. Also, physical paths should be specifically designed to connect to accelerators.
Mitigating Trombone effect
The design of the physical communication paths in the data centers can create a Trombone effect when a virtual path is designed where information has to traverse through a series of communication segments before reaching its destination in spite of physical proximity. This can adversely affect the latency. The data centers would need to be planned keeping in mind the need to support multipoint transmission. The design must balance hub and spoke topology, (which often is the underlying cause of the Trombone effect) with leaf and spine, and other topologies.
Efficient dynamic demand driven scaling
The data center must take into account the possibility of dynamically varying communication demands from consumers. Though NFV and SDN provide the capabilities to meet the same, they are limited by the physical infrastructure. The architecture of a CSP data center must plan for rapid augmentation of physical infrastructure. The DC should be easily augmentable with third-party cloud resources.
Virtualization technologies like NFV and SDN have enabled CSPs to leverage the concept of data centers from the enterprise world. However, the telco data centers serve a different purpose than the enterprise data centers and their design must address many concerns specific to the Communications industry. While the payload of traditional data centers is the application providing services to the consumers, the payload of telco data center is predominantly the transmission of information between SAPs. Hence, the design of telco DC is network-centric rather than compute- or storage-centric.
Considering the fact that 5G will support multiple types of communication services, the DC network should be designed keeping in mind all possible target communication needs between SAPs. The compute & storage are designed in the DC taking into consideration the type of VNFs that will need to be deployed. The distribution aspect of DC is based on economics as well as special needs for low latency services, and the endemic nature of some services.
Vinay Devadatta
Practice Head - Innovation & Industry Relations, Wipro Limited.
Vinay has extensive experience in the communications industry, working with equipment vendors, communication service providers, OSS product vendors, OSS solution providers and telco management standards bodies like TM Forum, ETSI NFV, NGMN NGCOR. He is responsible for leveraging external and internal innovation and creating new service lines at Wipro Technologies. His expertise is in the areas of OSS, NFV, Orchestration, automation, digital customer experience and 5G. He has led multiple cross organization projects exploring new concepts and futuristic Service Provider needs. He holds a B.Tech in Electronics & Communication Engineering and an M.Tech in Computer Science & Technology from IIT Roorkee.