Figure 1. Overbooking of Network Slice Resources

The Quest for New Revenue Models: Overbooking

Research work in the 5G area has pioneered novel revenue model solutions to address the aforementioned inefficiencies challenges on resource usage, e.g. [13-14]. Among them, Overbooking has been identified as one of the most promising ones.

5G leads mobile operators towards business models that, perhaps surprisingly, have a similar nature to successful yield management strategies popular in areas such as airline or hotel industries, and promise substantial gains in the revenue attained to mobile investments. In particular, in [3] the concept of slice overbooking is explored, i.e. accommodating the common practice in airline services of intentionally allocating more cargo than available capacity to the allocation of mobile network slices for 3rd-party services.

The challenge to adopt an orchestration system based upon the concept of slice overbooking is  threefold: (i) when doing overbooking, resource deficit (and thus violations of system-level agreements) may occur; and so, in order to maintain the incentives for 3rd-parties (users) to join the system, a balance between overbooking and potential service disruption must be taken care of; (ii) there is a need to untangle the coupling between resource reservation and slice admission control decisions (AC-RR), which is further compounded by the heterogeneous nature of the resources required to build a slice across the whole system; (iii) appropriate use of monitoring information is needed to be able to adapt to behavioral dynamics of 3rd-party services embedded in network slices.

Figure 2. Network Slicing Orchestration system (c.f. [16])

Overbooking in Practice: Leveraging on AI and Open-Source

A complete orchestration solution is presented in [16]. As shown in Fig. 2, the orchestration solution is designed to manage end-to-end slices across multiple heterogeneous domains of the mobile ecosystem, including Radio Access Network, Transport Network and Cloud Data Centers. Such a design is based on a hierarchical control plane that governs multiple domain controllers across a mobile system using ETSI-compliant interfaces and data models. The system embeds a control engine in charge of making (i) admission control and (ii) resource reservation decisions by exploiting monitoring and forecasting information while maximizing overall revenues. Such a maximization of business revenues falls into the category of yield management, a mainstream business theory that studies fare management, access control and resource allocation. In the airline industry, the problem is to decide, based on the number of seat reservations, whether to accept or reject new requests considering that passengers may cancel, or even be “no-shows”, prior to the flight departure. Thus, overbooking is performed with associated penalties determined by a penalty-cost function.

Analogously to the airline example, this solution design exploits the fact that users rarely consumes all the resources they request. This enables the allocation of more tenants than those presumably allowed by the leftover capacity thereby gaining additional revenue from the slice resources multiplexing (i.e., overbooking). Clearly, an overly aggressive strategy may lead to resource deficit, discouraging potential users to join the system. This is properly addressed by designing a proper penalty-cost function.

Thus, we can formulate our problem as an optimization problem that aims at minimizing the estimated penalty while, at the same time, maximizing the overall obtained reward. This formulation is subject to different system constraints that make the problem tractable and consistent, such as (i) guaranteeing that the aggregate amount of spectrum, network link bandwidth, computing capacity, memory capacity allocated to slices is never surpassed,  (ii) guaranteeing that accepted slices are allocated, at least, with the forecasted amount of resources and, at most, with the traffic requests, (iii) ensuring that all cloud resources (computing, memory, etc.) are allocated in the same data center.

Such a problem formulation can be mapped onto a Mixed Integer Linear Programming model, which can be solved with standard solvers (cplex) and algorithms (e.g. cutting-plane algorithms such as Benders decomposition method used in [16]).

Once the overbooking mechanism is grounded on an optimization formulation, provably-performing algorithms can achieve up to 3x revenue gains in several realistic scenarios built upon data from three real mobile operators. An experimental proof-of-concept validates the feasibility of implementing this approach with conventional mobile equipment on top of available open-source software, as shown in Fig. 3.

Figure 3. Network slicing overbooking in practice

Conclusion

While early 5G research and field-trials results are promising, we are far from fulfilling the holistic 5G vision. First 5G products will deliver mainly on the higher data rates promise. AI-based operations and managements solutions enabling the automation of 5G networks for a cost-efficient digitalization of industry verticals are still at its infancy and will require of a huge effort by the research community to reach commercial deployments during the next years.

 Future research challenges to be addressed will comprise enabling novel business models, opening the mobile network ecosystem to new players fostering increased innovation and achieving a security level that can be fully trusted by mission-critical infrastructures.

References

1. “Digital Transformation Initiative - Telecommunications Industry”, World Economic Forum White Paper, 2017
2. P.Caballero et al.: “Network slicing games: Enabling customization in multi-tenant networks”, IEEE INFOCOM, 2017.
3. P.Caballero et al.: “Network Slicing for Guaranteed Rate Services: Admission Control and Resource Allocation Games”, IEEE Transactions on Wireless Communications (TWC), 2018.
4. B.Han, V.Sciancalepore, D.Feng, X.Costa-Pérez, H.D.Schotten: “A Utility-Driven Multi-Queue Admission Control Solution for Network Slicing.”, IEEE INFOCOM, 2019.
5. P.Caballero et al.: “Network slicing games: Enabling customization in multi-tenant mobile networks”, IEEE/ACM Transactions on Networking (TON), 2019.
6. V.Sciancalepore et al.: “Mobile traffic forecasting for maximizing 5G network slicing resource utilization”, IEEE INFOCOM, 2017.
7. P.Caballero et al.: “Multi-Tenant Radio Access Network Slicing: Statistical Multiplexing of Spatial Loads”, IEEE/ACM Transactions on Networking (TON), 2017.
8. D.Bega et al.: “DeepCog: Cognitive Network Management in Sliced 5G Networks with Deep Learning”, IEEE INFOCOM, 2019.
9. V.Sciancalepore et al.: “STORNS: Stochastic Radio Access Network Slicing”, IEEE ICC, 2019.
10. V.Sciancalepore et al.: “RL-NSB: Reinforcement Learning-based 5G Network Slice Broker”, IEEE/ACM Transactions on Networking (TON), 2019.
11. C.Marquez et al.: “Resource Sharing Efficiency in Network Slicing”, IEEE Transactions on Network and Service Management (TNSM), 2019.
12. C.Marquez et al.: “How Should I Slice My Network?: A Multi-Service Empirical Evaluation of Resource Sharing Efficiency”, ACM MOBICOM, 2018
13. D.Bega et al.: “Optimising 5G infrastructure markets: The business of network slicing”, IEEE INFOCOM, 2017.
14. D.Bega et al.: “A Machine Learning approach to 5G Infrastructure Market optimization”, IEEE Transactions on Mobile Computing (TMC), 2019.
15. J.X.Salvat et al.: “Overbooking network slices through yield-driven end-to-end orchestration”, ACM CoNEXT. 2018.