Page 71 - ITU Journal Future and evolving technologies Volume 2 (2021), Issue 6 – Wireless communication systems in beyond 5G era

ITU Journal on Future and Evolving Technologies, Volume 2 (2021), Issue 6

into consideration bidders preferences, bids, bud‑ ent levels of granularity of resources in InP‑MVNO and
gets, and the demand at the seller’s side in the auc‑ MVNO‑slice users stages. Additionally, the authors ex‑
tion games. amined the transport and core networks of the 5G net‑
work. Similar to [28], an end‑to‑end 5G NS was exam‑
5. Finally, we present the performance evaluation of ined in [29]. The authors proposed a two‑level cen‑
the proposed framework via extensive Monte Carlo tralised resource allocation scheme termed as Upper‑
simulations. The proposed framework is compared tier First with Latency‑bounded over‑provision Preven‑
with schemes in the literature such as a genetic tion (UFLOP). The UFLOP prevents the overprovisioning
algorithm‑based dynamic resource allocation (GI‑ of resources between the InP and MVNOs while meeting
LARE) and the non‑dynamic approach of Static‑ the MVNOs’ Service Level Agreement (SLA) and slice use
Slicing (SS) schemes. cases’ latency requirements.
In [30, 31], the authors categorised slice users into Guar‑
1.2 Article organisation anteed Bit Rate (GBR) and non‑Guaranteed Bit Rate (non‑
GBR) users. In [30], the authors proposed a Markov‑
The rest of the article is structured as follows. We de‑
vote Section 2 to the discussion of related literature. In based resource allocation for a multi‑slice multi‑tenant
Section 3, we focus on the system model. Section 4 network. The proposed Markov‑based algorithm was
presents the multi‑domain multi‑tenant auction games able to characterise and model the stochastic behaviour
framework. Further, Section 5 describes the resource al‑ of slice use case requirements and channel behaviour.
location framework in an M‑TTSD 5G NS network. Then Moreover, the authors [31] proposed an MVNO virtual
in Section 6, we formulate the M‑TTSD resource allocation resource allocation architecture to accurately predict the
problem. Section 7 details the reformulation steps and bandwidth requirement of different slice use cases, re‑
proposed solutions. In Section 8, the complexity analysis spectively. The proposed architecture relies on the Long
of the proposed solutions is discussed. Then, in Section 9, Short‑Term Memory (LSTM) neural network for predic‑
we present the numerical results of the performance eval‑ tion.
uation. Finally, conclusions are drawn in Section 10. An iterative double‑auction mechanism for a multi‑tenant
multi‑domain SDN‑based network was proposed by the
authors in [32] to maximise the utility of the respective
2. RELATED WORK
MVNOs.
Our work builds on the Fisher Market (FM), which is one Whereas the authors in [7] enumerated different business
of the most widely known models in mathematical eco‑ models for wireless network virtualisation in 5G and be‑
nomics [11, 12]. In FM, a set of buyers with limited bud‑ yond era. Additionally, the authors did not consider the
gets (of no intrinsic value) to purchase divisible goods Radio Access Network (RAN) and multi‑slice characteri‑
from a set of sellers with the sole aim of maximising their sation of the network with its diverse QoS requirement.
respective non‑negative utility [13, 14, 15]. In the same In [33, 34, 35], a two‑level stackelberg game and win‑
vein, we liken InPs, MVNOs, SPs, and slice users to sellers ner determination models for dynamic pricing were pro‑
and buyers in the M‑TTSD network. posed by the authors. The authors considered a tradi‑
The works in [16, 17, 18, 19] proposed optimisation tional multi‑tenancy scenario with a single seller (InP)
frameworks for maximising the utility of slice users in and multiple buyers (MVNOs). In [34], a hierarchical auc‑
a single‑domain multi‑tenant network. The proposed tion model was examined. In [36], the authors considered
frameworks employed static‑slicing schemes. Network a hierarchical multi‑tenant network comprising InPs, an
economics variables such as the budgets of the respective MVNO, and SPs. The scenario is such that an SP can only
network players were not considered. bid for resources from one MVNO, while the MVNO can
Contrary to the aforementioned works in [16, 17, 18, 19], network resources from multiple InPs.
the authors [20, 21, 22, 23] adopted the network eco‑ Similar to [36], the authors in [37] adapted the Non‑
nomics variables (such as budget, auctions) in their pro‑ Orthogonal Multiple Access (NOMA) technique to allocate
posed schemes for ef icient resource allocation. Auction resources of the MVNOs and SPs. A Tchebycheff technique
game‑based schemes were proposed for resource alloca‑ was employed in solving the Multi‑Objective resource al‑
tion between an InP (seller) and MVNOs (buyers). location Optimisation Problem (MOOP). The authors in
In [24], the authors proposed a dynamic virtual resource [38, 39] employed a MOOP technique to the single domain
allocation framework premised on the widely known bio‑ heterogeneous networks.
logical population model called Lotka‑Volterra [46]. The A deep learning‑based caching and leasing framework
virtual resources of the multiple InPs are centralised and was proposed in [40] for a traditional single‑domain
then allocated to MVNOs. Similar to [24], the authors multi‑tenant network comprising an InP and a set of
in [25, 26, 27] proposed dynamic resource allocation MVNOs. The algorithm predicts the resource leasing pat‑
schemes which are centrally controlled in heterogeneous terns of respective MVNOs for pro it maximisation.
networks. In [41, 42], the authors proposed a weighted proportional
A two‑stage resource allocation scheme for a multi‑tenant allocation scheme that allocates power in a single‑tier tra‑
slice network was proposed in [28] to address the differ‑ ditional multi‑tenancy virtual network. A bidding strat‑

© International Telecommunication Union, 2021 59