Cost minimization in placing service chains for virtualized network functions

Abstract

Network function virtualization (NFV) places network functions onto the virtual machines (VMs) of physical machines (PMs) located in data centers. In practice, a data flow may pass through multiple network functions, which collectively form a service chain across multiple VMs residing on the same or different PMs. Given a set of service chains, network operators have two options for placing them: (a) minimizing the number of VMs and PMs so as to reduce the server rental cost or (b) placing VMs running network functions belonging to the same service chain on the same or nearby PMs so as to reduce the network delay. In determining the optimal service chain placement, operators face the problem of minimizing the server cost while still satisfying the end-to-end delay constraint. The present study proposes an optimization model to solve this problem using a nonlinear programming (NLP) approach. The proposed model is used to explore various operational problems in the service chain placement field. The results suggest that the optimal cost ratio for PMs with high, hybrid, and low capacity, respectively, is equal to 4:2:1. Meanwhile, the maximum operating utilization rate should be limited to 55% in order to minimize the rental cost. Regarding quality of service (QoS) relaxation, the server cost reduces by 20%, 30%, and 32% as the end-to-end delay constraint is relaxed from 40 to 60, 80, and 100 ms, respectively. For the server location, the cost decreases by 25% when the high-capacity PMs are decentralized rather than centralized. Finally, the cost reduces by 40% as the repetition rate in the service chain increases from 0 to 2.A heuristic algorithm, designated as common sub chain placement first (CPF), is proposed to solve the service chain placement problem for large-scale problems (eg, 256 PMs). It is shown that the proposed algorithm reduces the solution time by up to 86% compared with the NLP optimization model, with an accuracy reduction of just 8%. KEYWORDSnetwork function placement, network function virtualization, nonlinear programming, service chains Int J Commun Syst. 2020;33:e4222.wileyonlinelibrary.com/journal/dac

AI based Service Management for 6G Green Communications

“…Another critical paradigm is the utilization of Artificial Intelligence (AI) techniques to provide context-aware information transmissions and personal-customized services, as well as realize the automatic network management [1], [10], [11]. The growing ICT infrastructure, exploding data, and the increasingly complex network management will result in surging energy consumption, which poses a great challenge for the network operators [12], [13]. Data analysis shows that the ICT sector may cost more than 20% of the total electricity [5] as in Fig.…”

Robust and adaptive decision-making: a reinforcement learning perspective

“…Environments CD CommNet [81] Predator Prey (PP) [95] LC TarMAC [12] Traffic Junction (TJ) [81] CC NDQ [15] StarCraft II (SCII) [96] Table 5.1: The chosen algorithms and environments.…”

“…Following the above taxonomy, we select some representative state-of-the-art algorithms to perform evaluation 3 . As in Table 5.1, we select CommNet [81], Tar-MAC [12] and NDQ [15] for CD, LC and CC respectively. CommNet [81] is a simple yet effective multi-agent communicative algorithm where incoming messages of each agent are generated by averaging the messages sent by all agents in the last time step.…”

“…Then a weighted sum over all incoming messages is performed to determine the message an agent will receive for decentralized execution. NDQ [15] achieves nearly decomposable Q-functions via communication minimization. Specifically, it trains the communication model by maximizing mutual information between agents' action selection and communication messages while minimizing the entropy of messages between agents.…”

“…• Communication with cost or limited bandwidth (CC): Agents should avoid sending redundant messages because communication in real world is costly and communication channels often have a limited bandwidth [15,16].…”