There are few debating that disaggregation of the RAN along with adoption of programmability will bring new monetization opportunities to the RAN. However, one question that is being debated is: in the context of network transformation, how soon can we start building successful business models that rely on open RAN?
During the recent O-RAN Alliance NA Spring 2022 PlugFest, VMware and Intel demonstrated that the answer is now.
To enable Communication Services Providers (CSPs) looking to RAN slicing as a vehicle to guarantee SLAs, we proudly presented our dynamic RAN slice resource allocation solution for SLA assurance. This technological concept uses RAN programmability to monitor and dynamically control resources available in RAN slices to meet negotiated customer SLAs. Our simple deployment, which required no major changes in the network, consisted of the Intel® FlexRAN™ Reference Architecture with VMware Distributed RIC and a RAN slice resource allocation xApp from VMware.
This demonstration showed immediate promise for business scenarios.
The Business Scenario
Imagine that you are an engineer at a CSP and that one of your enterprise customers wants to launch a new service. Connecting through your BSS, this enterprise customer negotiates a Service Level Agreement (SLA). To fulfill this agreement, you decide to create a RAN slice composed of multiple cells that will be governed by conformance to the SLA.
So far, so good, but:
- How do you assure this SLA?
- How do you assure the SLA while optimizing resource utilization in the slice?
These are the questions that VMware and Intel’s demo, “Dynamic RAN slice resources allocation solution for SLA assurance”, answered.
Let’s review how.
The Technical Solution
The technological concept uses RAN programmability activated through the Near-RT RIC to monitor and dynamically control resources available in RAN slices to meet the negotiated customer SLAs.
The result of this intervention is mimicked in Figure 2, where you can see how the xApp and VMware Distributed RIC (VMware Near-RT RIC) are constantly changing allocated PRBs on each cell to guarantee the per-UE minimum throughput (the negotiated SLA in this case). The allocation ensures that all available radio resources are used in an optimal fashion.
This solution is composed of the following key elements (Figure 3):
- VMware Distributed RIC –a near real-time RIC integrating Intel® FlexRAN™ software and APIs and deployed on latency-sensitive Kubernetes worker node
- A RAN slice PRB allocation xApp from VMware –also deployed on a latency-sensitive Kubernetes worker node and onboarded through VMware RIC SDK.
- The VMware RIC SDK – software development kit containing libraries and APIs to aid with the onboarding of xApps and simplify interoperability
- Intel® FlexRAN™ software-based O-DU or any other O-DU running
All these elements work together harmoniously to produce the logic that guarantees SLAs and optimizes resource utilization. Here is how it works:
As UEs connect to the slice and start generating traffic, the O-DUs use the embedded Intel FlexRAN libraries to report cell and user performance measurements to the VMware Distributed RIC and xApp. Using the E2 interface (E2SM-KPM) on the northbound side, the O-DUs submit:
- User RF condition and achieved throughput
- Cell-level PRB utilization
The VMware xApp receives these measurements from the O-DUs and uses them, along with the negotiated SLA and overall resource quota per slice, as inputs for the algorithm that dynamically calculates the new PRB allocation values per cell.
Then, in near real-time, the xApp and VMware Distributed RIC access the embedded Intel FlexRAN RAN Control (E2SM-RC) libraries to issue a set of commands ordering each O-DU to modify cell scheduling to adjust the PRB allocation to the UEs belonging to the slice in order to achieve the per-user minimum throughput SLA.
Each O-DU handles these asynchronous commands and executes them, effectively enforcing the negotiated SLA.
The O-RAN Alliance PlugFest Demonstration
Working in collaboration with Rutgers University, we deployed our solution in its WINLAB (https://www.winlab.rutgers.edu/), located in North Brunswick, New Jersey.
The testbed setup (seen in Figure 4) involved one instance of VMware Distributed RIC and one instance of the demo xApp from VMware integrated with the E2-compliant Intel FlexRAN O-DU solution and Intel UE simulators.
The demonstration conditions were the following:
- We defined a RAN slice composed of three cells with varying RF conditions per cell
- Then, we loaded the Intel UE simulator with a distribution of UEs – 1 UE, 2 UEs and 4 UEs per cell, with varying RF conditions per UE
- The carrier bandwidth on each cell was 100 MHz
- We set up initial equal distribution of scheduling resources across all cells
- And finally, we set up an SLA of minimum UE throughput of 30Mbps
The Results
To demonstrate the impact of our solution, first we showed what would be the performance of the UEs without any RIC or xApp intervention.
We activated our simulated UEs and used a Grafana dashboard to display the measurements that were being reported to the VMware Distributed RIC through the Intel FlexRAN E2SM-KPM libraries. The results are shown in Figure 5 below. The bottom chart in the figure shows the PRBs allocated to the slice users per cell, which is the same across the three cells due to lack of intervention by the xApp and VMware Distributed RIC. Due to the variable user loading across the cells, this PRB allocation strategy results in a highly variable user throughput with some slice users not being able to meet their SLA’s. For example, in Figure 5, you can see how the UE’s throughput of Cell 3 (the blue line) was 24 Mbps (below the 30 Mbps minimum SLA) while some of the other UEs benefited from service levels above the negotiated level.
Next, we show the results we obtained after deploying the RAN slice PRB allocation xApp from VMware with enforced SLA:
First, we showed VMware Distributed RIC and the RAN sSlice PRB allocation xApp using the integrated Intel FlexRANTM E2SM-KPM libraries to monitor the performance of each cell composing the RAN slice and each UE. Then, based on instant traffic demands, we showed how the xApp activated the algorithm that calculated the PRB allocation corrections to guarantee the SLA and how these corrections were communicated to the O-DUs through the RIC in the form of E2SM-RC control commands. As you can see in Figure 6, these commands produced changes in the PRB allocations per cell in the slice (bottom chart). These dynamic changes in PRB allocation in turn produced dynamic changes in the per-UE throughput (top chart).
As shown in the chart, the reported minimum throughput for the slice UEs in all three cells always remained above the 30Mbps mark. Moreover, we could also see that the minimum UE throughput across the cells were converging to a similar performance value thanks to a more intelligent use of the available resources.
The results were clear: thanks to VMware Distributed RIC and the RAN slice PRB allocation xApp from VMware, all UEs throughput met the minimum target as per SLA, and RAN slice resources were used optimally when and where needed.
The Benefits
From a technical perspective, we have demonstrated that deploying our VMware Distributed RIC with the Intel® FlexRAN™ Reference Architecture immediately delivers multiple technical advantages. To give just a few examples:
- Per user, cell and slice performance visibility
- Control over network and user-level performance
- Maximization of radio resources utilization
These key technical ingredients offered by our solution have immediate business applicability for CSPs: by assuring RAN slice SLAs, this solution could be the cornerstone that CSPs can build upon to customize the virtual enterprise network based on customer’s business preferences and to differentiate service offerings through SLAs.
