What is RAN slicing
Microsoft Radio access network (RAN) slicing has generated great excitement in the scientific and marketing worlds. One of the key new aspects of 5G networks is RAN slicing, which enables differentiated services and opens up new revenue streams for operators and users alike.
RAN slicing 5G 3GPP
The slice method is specified in the Third Generation Partnership Project (3GPP) specifications, however the implementation of the slices is left up to the individual specifications. Furthermore, Microsoft hasn’t seen many real-world, production-level RAN slicing implementations; this could be due to the complexity of the 5G business roll-out. They like to list a few of the latest findings from Microsoft’s research on Microsoft RAN slicing that will facilitate operators’ adoption of it with Microsoft Azure.
Guaranteed service using RAN slicing
Reliable network throughput and latency are necessary for latency-sensitive mobile applications including cloud robotics, Microsoft Teams video conferencing, Xbox Cloud Gaming, Microsoft Mixed Reality, and remote telemedicine. The 3GPP specifications established network slicing, a virtualization fundamental that enables an operator to run numerous distinct virtual networks, or slices, built on top of a single physical network, in recognition of this need for next-generation mobile apps. Since the last-mile wireless link frequently acts as a barrier for mobile apps, Microsoft RAN slicing is especially interesting for service assurance.
The Technical Issue
The resource allocation policy of a network should ideally be able to be adjusted by a network operator to meet the unique connectivity needs of any application that subscribes. However, in practice, base station schedulers typically optimise for coarse metrics, like the total throughput attained by a group of apps or the aggregate throughput attained by the base station. The issue is that neither of these approaches guarantees sufficient performance for every networked application.
A group of users or a group of apps with comparable connectivity needs can be supported by a network slice. To offer varied connectivity, operators might allocate resources, such as physical resource blocks (PRBs), among the slices in the RAN slicing.
In order to ensure slice-level service assurance through service-level agreements (SLAs), existing techniques divide up PRBs among various slices. However, as they previously stated, service assurance needs to be given at the application level in order to accomplish the desired results, where programmes attain the network performance they need. Current methods are insufficient to allow operators to offer this crucial feature.
Slice-level service assurance does not provide latency and throughput for every app within the slice because channel conditions can vary greatly amongst users in the same slice. The fact that programmes join and exit the network asynchronously further complicates optimization. To satisfy the needs of every app in a slice, Microsoft needs to provide app-level service assurance. Microsoft recognised and tackled the following two issues in order to achieve this:
State-space intricacy
Previous methods track a state space comprising aggregate slice-level statistics, such as the observed slice throughput and the average channel quality of all users in a slice, to offer slice-level service assurance. One may treat each app as a slice to extend these approaches to support requirements at the app level. The issue is that this broadens the state space to encompass each app’s observed latency, observed throughput, and channel quality.
All possible values for the tracked variables are included in the resulting state space, which expands swiftly. Searching through this state space to find a PRB allocation that complies with the apps SLA causes an unsolvable optimization issue in real-world deployments where hundreds of apps need to be hosted on the network.
Figuring out the availability of resources
Operators usually use admission controllers, which accept or deny incoming apps based on a policy, to compute bandwidth allotment for slices. The policy may be determined by limits on fairness, preferences for slice monetization, or other goals. A great deal of research has been done on admission control algorithms. Basically, what operators need is a means to assess if the RAN slicing has enough resources to support an incoming app’s SLAs without adversely affecting those of apps that have previously been admitted. Regrettably, while previous methods computed necessary PRBs to enable slice-level SLAs, they are hard to modify. Once more, considering each app as a slice is not possible due to the state-space complexity.
Examine Microsoft’s RAN-slicing mechanism
Microsoft has created a radio resource scheduler that satisfies latency and throughput SLAs for distinct programmes running over cellular networks. The Microsoft system groups applications into network slices based on comparable SLA demands. It leverages traditional schedulers, which maximise throughput at the base station by calculating resource schedules for each slice in a way that meets the needs of each application. Apps communicate their minimum throughput and maximum delay requirements to the operator under this paradigm. The Microsoft system then computes and allocates the PRBs needed by each slice in order to meet these SLAs via the shared wireless medium on behalf of the operator.
The following methods are used by Microsoft System to overcome the difficulties in allowing app-level service assurance in a wireless environment:
- Microsoft controls the intricacy of the search area and separates the control policy from the network model. Microsoft accomplishes this by expressing bandwidth allocation that complies with SLAs as a model predictive control (MPC) problem. Sequential decision-making problems with a shifting look-ahead horizon are well solved by MPC. It separates a predictor, which expressly represents environmental uncertainty, from a controller, which resolves a standard optimisation problem.
- Microsoft forecasts all state-space variables, including the wireless channel that each app uses, using standalone predictors. The Microsoft system uses these forecasts to feed a control algorithm, which uses the anticipated state to generate a series of future bandwidths for each slice.
- Because Microsoft notes that app throughput and latency vary monotonically with the number of PRBs, Microsoft reduces complexity by letting Microsoft control algorithm effectively minimise the search space of feasible bandwidth allocations.
- Microsoft created a series of deep neural networks to estimate the distribution of necessary PRBs in order to forecast the availability of RAN resources. Microsoft uses offline simulations of their control algorithm to train these neural networks, which are subsequently used to forecast resource availability in real time.
Microsoft bases its base bandwidth (PRB) allocation on anticipated channel circumstances at a high level. Microsoft believes that packet loss will be reduced and that the PRB allocation will match the request made by the application when the signal to noise ratio (SNR) is high. Packet loss will be higher when SNR is low, hence PRB allocation will be increased to make up for it.
The Microsoft system exposes a primitive that determines whether there is enough bandwidth to meet the needs of an incoming application in order to assist the admittance controller. The wonderful thing about this is that the operator can apply their monetization rules independently because the admission control policies are not dependent on the availability of bandwidth.
5G RAM slicing
The above concepts are realised via a system that is Microsoft O-RAN compliant. Microsoft’s production-class, end-to-end 5G platform now includes the Microsoft RAN slicing system. Microsoft added hooks to the vRAN distributed unit’s various modules to allow for dynamic slice bandwidth adjustment without sacrificing real-time performance.
To accommodate various traffic types and enterprise policies, the operator can setup its Microsoft RAN slicing with a set of slices. For instance, it can set up different slices for Xbox Cloud Gaming sessions and Microsoft Teams sessions. Measured as a ratio of the violation of the app’s request, Microsoft considerably reduces SLA breaches as compared to a slice-level service assurance scheduler. With the help of the Microsoft system, operators may overcome the significant difficulty of giving apps reliable network performance. In this manner, a production-class vRAN can be integrated with app-level service assurance.
Find answers that give developers more control
It is Microsoft’s goal to bring programmable networks to reality. Microsoft considers this to be an essential, core skill for developers to create services and write apps that are far superior to the apps that are available now. Slicing the network Microsoft RAN is a crucial stage in this process. Microsoft is able to handle time-sensitive and secure applications that need consistent, continuous bandwidth with Microsoft RAN slicing. As a result, operators will be able to offer a wide range of innovative and alluring network service features to developers of next-generation applications while maintaining operational efficiency.
