Getting the Most Out of Your VNFs: Flexible Assignment of Service Priorities in 5G

Abstract

Through their computational and forwarding capabilities, 5G networks can support multiple vertical services. Such services may include several common virtual (network) functions (VNFs), which could be shared to increase resource efficiency. In this paper, we focus on the seldom studied VNFsharing problem, and decide (i) whether sharing a VNF instance is possible/beneficial or not, (ii) how to scale virtual machines hosting the VNFs to share, and (iii) the priorities of the different services sharing the same VNF. These decisions are made with the aim to minimize the mobile operator's costs while meeting the verticals' performance requirements. Importantly, we show that the aforementioned priorities should not be determined a priori on a per-service basis, rather they should change across VNFs since such additional flexibility allows for more efficient solutions. We then present an effective methodology called FlexShare, enabling near-optimal VNF-sharing decisions in polynomial time. Our performance evaluation, using real-world VNF graphs, confirms the effectiveness of our approach, which consistently outperforms baseline solutions using per-service priorities.

OKpi: All-KPI Network Slicing Through Efficient Resource Allocation

“…In the context of MEC, some works tackle tasks different from sheer data processing; as an example, [11], [16] aim at jointly optimizing computation and caching offloading between cloud-based and MEC-based infrastructure. Others focus on additional decisions that can be made in slicing scenarios, e.g., priority assignment in [3]. A body of works considers incremental deployment, i.e., service requests arriving at different times: in this case, it is possible to share existing VNF instances [3], [7], [17], augment routing paths instead of computing them from scratch [7], [17], and minimize the difference between current and future network configuration [17].…”

“…Others focus on additional decisions that can be made in slicing scenarios, e.g., priority assignment in [3]. A body of works considers incremental deployment, i.e., service requests arriving at different times: in this case, it is possible to share existing VNF instances [3], [7], [17], augment routing paths instead of computing them from scratch [7], [17], and minimize the difference between current and future network configuration [17]. Among the few works tackling non-functional requirements, [2] performs resilient VNF placement, to achieve robustness to equipment failures.…”

“…Furthermore, existing studies have limited or no support for several specific aspects of slicing-based networks, including (i) the opportunity to reuse existing VNF instances ( [3] reuses VNFs within a single data center, while [7] only focuses on cost), (ii) the possibility of combining cloud-based and MECbased services (with the exception of [8], which however only deals with caching), and (iii) the need to make decisions on how to place and connect VNFs, thus jointly addressing VNF placement and data routing ( [9]- [12] do so, but without considering PoAs or VNF re-usage, and under some limiting assumptions, e.g., on the number of VNF instances).…”

Function Splitting, Isolation, and Placement Trade-Offs in Network Slicing

“…Secondly, we show by numerical simulations the impact of taking into full and partial consideration these peculiar novel technical constraints. Even though several works partially cover the network slice design problem [12], [13] and related sub-problems, such as functional split mode selection [14]- [19], network slicing with VNF sharing [20]- [22], and network slicing with VNF scaling [23]- [27], no attention has been given to address jointly all aforementioned aspects in order to design network slices and understand the impact of mapping, sharing, and split policies on both virtual and physical networks. This paper is an extension of [1], differing in the following aspects: (i) we model new variants and extensions of the problem, and (ii) we provide additional analyses regarding the impact of each proposed variant on the network.…”

On the impact of novel function mappings, sharing policies, and split settings in network slice design

“…Also, our model is compliant with technical specifications published by the 3rd Generation Partnership Project (3GPP) [2], [9], [10]. Even though several works partially cover the network slice design problem [11], [12] and related sub-problems, such as functional split mode selection [13]- [15], network slicing with VNF sharing [16]- [18], and network slicing with VNF scaling [19]- [23], no attention has been given to address jointly all aforementioned aspects in order to design network slices and understand the impact of mapping, sharing and split policies.…”