The marketing and scientific communities are excited about Radio Access Network (RAN) slicing. RAN slicing is one of the key new features of 5G networks. This enables differentiated services, enabling new features for customers and network monetization opportunities for carriers. The 3rd Generation Partnership Project (3GPP) specification defines a slicing mechanism but does not say anything about how slicing is implemented. Also, perhaps due to the complexity of 5G business deployment, he doesn't see many examples of RAN slicing actually being implemented at the production level. We conducted research and produced new results related to RAN slicing. I would like to enumerate some results that allow an operator to easily use RAN slicing in Microsoft Azure.
Azure for operators
Modernize and monetize your network
Service Assurance with RAN Slicing
Latency-sensitive mobile applications such as Xbox Cloud Gaming, Microsoft Teams video conferencing, Microsoft Mixed Reality, remote telemedicine, and cloud robotics require predictable network throughput and latency. The 3GPP specifications recognized this requirement for next-generation mobile apps and introduced network slicing. Network slicing is a virtualization primitive that allows operators to run multiple differentiated virtual networks, called slices, layered on top of a single physical network. RAN slicing is especially important for service assurance because the last mile radio link is often the bottleneck for mobile apps.
technical issues
Ideally, network operators should be able to configure the network's resource allocation policies to meet the specific connectivity requirements of each subscribing application. However, in the real world, typical base station schedulers optimize for coarse-grained metrics, such as the aggregate throughput at the base station or the aggregate throughput achieved by a bundle of applications. The problem is that none of these methods can guarantee proper performance for each application connected to the network.
A network slice can support a set of users or a set of applications with similar connectivity requirements. An operator can distribute resources such as physical resource blocks (PRBs) across slices within her RAN to provide differentiated connectivity.
Existing approaches allocate PRBs to different slices and guarantee slice-level service guarantees through service level agreements (SLAs). However, as mentioned earlier, service guarantees must be provided at the application level for apps to realize the expected benefits of achieving the required network performance. Existing approaches fall short of providing operators with this critical capability. Slice-level service guarantees do not guarantee throughput and latency for each app within a slice because different users within the same slice can experience significantly different channel conditions. Apps also join and leave the network asynchronously, making optimization difficult. App-level service guarantees are required to meet the requirements of each app within a slice. To achieve this, we identified and addressed two challenges:
- state space complexity
Traditional approaches provide slice-level service guarantees by tracking a state space consisting of aggregated slice-level statistics, such as the average channel quality of all users within a slice and the observed slice throughput. Masu. To extend these methods to support app-level requirements, you can treat each app as a slice. The problem is that doing so expands the state space to include the channel quality, observed throughput, and observed delay experienced by each app. The resulting state space consists of all the possible values of the tracked variables and grows rapidly, so searching this state space to determine PRB assignments that comply with the app's SLAs will cause the network to The actual deployment in use presents intractable optimization problems. It needs to support hundreds of apps. - Determining resource availability
To calculate bandwidth allocation for a slice, operators typically run an admission controller that allows or denies incoming apps according to some policy. Policies may depend on slice monetization settings, fairness constraints, or other objectives. Admission control algorithms have been widely studied. Essentially, a carrier needs a way to determine whether her RAN has the resources to accommodate her SLA for an incoming app without negatively impacting her SLA for apps that are already allowed. That's what I mean. Unfortunately, traditional approaches are difficult to adapt because they calculate her PRB needed to support her SLA at the slice level. Again, the complexity of the state space makes it impossible to treat each app as a slice.
Explore Microsoft's RAN slicing system
We designed and developed a radio resource scheduler that meets throughput and latency SLAs for individual apps running on cellular networks. Our system bundles apps with similar SLA requests into network slices. It leverages a traditional scheduler that maximizes base station throughput by calculating the resource schedule for each slice in a way that meets the requirements of each app. In this model, the app expresses network requirements to the operator in the following format: minimum throughput and maximum delay. On behalf of the operator, our system satisfies these his SLAs over the shared wireless medium by calculating and allocating the required PRBs to each slice.
Our system addresses the challenges of achieving app-level service guarantees in wireless environments by applying the following techniques:
- We manage the complexity of the search space and separate the network model and control policy. This is achieved by formulating SLA-compliant bandwidth allocation as a model predictive control (MPC) problem. MPC excels at solving continuous decision-making problems over moving lookahead periods. This separates the controller, which solves classical optimization problems, from the predictor, which explicitly models the uncertainty in the environment.
- Use standalone predictors to predict each state-space variable, such as the wireless channels, that each app experiences. Our system then inputs these predictions into a control algorithm that calculates a sequence of future bandwidths for each slice based on the predicted state.
- Because we note that app throughput and latency vary monotonically with the number of PRBs, we reduce complexity by allowing the control algorithm to efficiently prune the search space of possible bandwidth allocations. Reduce.
- Predict RAN resource availability by designing a family of deep neural networks that predict the distribution of required PRBs. Train these neural networks offline with simulations of control algorithms and apply them to predict resource availability in real time.
At a high level, bandwidth (PRB) allocation is based on predicted channel conditions. If the signal-to-noise ratio (SNR) is high, packet loss is low and the PRB allocation is likely to match what the app requested. When the SNR is low, the packet loss is high and the PRB allocation is high to compensate. To assist the admission controller, our system exposes a primitive that estimates whether there is enough bandwidth to accommodate the requirements of the receiving app. The good thing about this is that the admission control policy does not depend on bandwidth availability, so carriers can implement their own monetization policies.
Our O-RAN compatible system realizes the above idea. We have implemented his RAN slicing system on a production-class end-to-end 5G platform. We implemented hooks on different modules of the vRAN distribution unit to dynamically control slice bandwidth without compromising real-time performance.
Operators can use sets of slices to configure the RAN according to different traffic types and enterprise policies. For example, you can set up separate slices for Microsoft Teams sessions and Xbox Cloud Gaming sessions. Compared to slice-level service guarantee schedulers, it significantly reduces SLA violations, measured as the ratio of app request violations. Our system enables carriers to solve the critical challenge of providing predictable network performance for apps. In this way, app-level service guarantees can be built into production-class vRAN.
Discover solutions that empower developers
Microsoft is passionate about making programmable networks a reality. We believe this is the fundamental functionality that developers need to create applications and build services that are significantly better than their current applications. Network RAN slicing is an important step in this effort. RAN slicing can be used to support secure, time-critical applications that require sustained and predictable bandwidth. This enables carriers to offer many new and exciting network service features with increased operational efficiency for next-generation application developers.
RAN slicing is a great idea and we're making it happen. We hope that various RAN vendors will incorporate these ideas as they integrate with Microsoft Azure Operator Nexus. Further technical details of what I wrote can be found in my recently published paper, “Application-Level Service Assurance with 5G RAN Slicing.”
