Solving the Multi-Objective Problem of IoT Service Placement in Fog Computing Using Cuckoo Search Algorithm

Abstract

The Internet of Things (IoT) has led to the proliferation of networked computing devices in the public, commercial, and private sectors. These devices often demand real-time computational resources, which are rather challenging to provide by the conventional cloud computing paradigm. In this regard, the emerging fog computing subtly solves this problem by providing more agile access to local storage and computational resources. Fog computing is a novel computational paradigm that provides resources in the proximity of IoT devices in such a way that all fog cells are located at the edge of the network. Although the theoretical foundations of fog computing have already been established, the problem of services placement for mapping IoT applications to fog cells aiming at maximizing the utilization of fog resources and improving Quality of Service (QoS) is still challenging. In order to fill this gap, we have developed a conceptual computing framework based on combination of cloud-fog by introducing the cloud-fog control middleware that manages service requests to meet some constraints. In this framework, the fog service placement problem is modeled as a multi-objective optimization problem that considers the heterogeneity of resources and applications based on QoS requirements. Finally, we propose an evolutionary algorithm based on the cuckoo search to solve the problem. The simulation results show that the proposed approach provides better performance compared to its counterparts in terms of various metrics such as fog utilization, energy consumption, number of performed services, response time, and service delay.

Appendices

Appendix 1

The abbreviations used in the paper are as follows. IoT: Internet of Things, QoS: Quality of Service, FC: Fog Computing, FSPP: Fog Service Placement Problem, CFCM: Cloud-Fog Control Middleware, MADE-k: Monitor-Analyze-Decision-Execute and knowledge base, CSA: Cuckoo Search Algorithm, FOCN: Fog Orchestration Control Node, NNFC: Nearest Neighbor Fog Colony, SLA: Service Level Agreement, SLATAH: SLA Violation Time per Active Host, PDM: Performance Degradation caused by Migration, SLAV:SLA Violation, FU: Fog Utilization, ELR: Egg Laying Radius, GWO: Gray Wolf Optimization, PSO: Particle Swarm Optimization, ODMA: Open-source Development Model Algorithm, BTU: Billing Time Unit.

Appendix 2

The notations and definitions of the variables applied in the fog service placement problem is as follows. $N$: Total number of colonies, $F^{i}$: The $i$-th fog orchestration control node, $Res\left( {F^{i} } \right)$: Subordinated fog cells in the $i$-th fog colony, $f^{j}$: The $j$-th fog cell, $c^{i}$: Number of subordinated fog cells in $F^{i}$, $d^{{f^{j} }}$: Communication link delay between $F^{i}$ and $f^{j}$, $d^{N}$: Communication link delay between $F^{i}$ and neighboring fog colony, $d^{R}$: Communication link delay between $F^{i}$ and cloud, $U_{{F^{i} }}$: Utilization CPU capacity of $F^{i}$, $U_{{f^{j} }}$: Utilization CPU capacity of $f^{j}$, $U_{{a_{l} }}$: The requested CPU of $l$-th service, $M_{{F^{i} }}$: Utilization RAM capacity of $F^{i}$, $M_{{f^{j} }}$: Utilization RAM capacity of $f^{j}$, $M_{{a_{l} }}$: The requested RAM of $l$-th service, $S_{{F^{i} }}$: Utilization storage capacity of $F^{i}$, $S_{{f^{j} }}$: Utilization storage capacity of $f^{j}$, $S_{{a_{l} }}$: The requested storage of $l$-th service, $AP$: Set of applications to be executed, $A_{k}$: The $k$-th application, $m$: Total number of applications, $r^{k}$: Number of services in $A_{k}$, $a_{l}$: The $l$-th service of an application, $D_{{A_{k} }}$: deadline of $A_{k}$, $D_{{a_{l} }}$: deadline of $a_{l}$, $RT_{{A_{k} }}$: Response time of $A_{k}$, $\varphi$: Percentage of free resources in fog cell, $m_{{A_{k} }}$: Makespan of $A_{k}$ application, $m_{{a_{l} }}$: Makespan of $a_{l}$, $d^{{A_{k} }}$: Total delay of $A_{k}$, $w_{{A_{k} }}$: Deployment time of $A_{k}$, $\rho$: The distance between two time periods of placement, $\tau$: Time unit postponed application deployment, $TS_{{f^{j} }}$: The time during which CPU is fully utilized by $f^{j}$, $TA_{{f^{j} }}$: The time during which $f^{j}$ state is active, $CD_{{F^{i} }}$: The performance degradation of $F^{i}$, $CR_{{F^{i} }}$: Total requested CPU capacity of $F^{i}$, $W_{T}$: Thin application energy in working state, $W_{F}$: Fat application energy in working state, $I_{T}$: Thin application energy in idle state, $I_{F}$: Fat application energy in idle state, $U_{l}$: Number of tasks executed for $a_{l}$, $P_{min}$: Minimum energy consumption in active state, $P_{max}$: Energy consumption at the peak load, $BW$: Bandwidth, $cv_{\alpha ,\beta }$: Data size between $t_{\alpha }$ and $t_{\beta }$ service requests, $CP_{{f^{j} }}$: The communication unit price of $f^{j}$, $PP_{{f^{j} }}$: The processing unit price of $f^{j}$, $C$: Total number of fog cells, $K$: Total number of IoT services.

Appendix 3

The notations and definitions of the variables applied in the cuckoo search algorithm is as follows. $\mu$: Location parameter, $c$: Scale parameter, $\Gamma$: Gamma function, $N_{P}$: Population size in CSA, $\overline{N}$: Pareto archive size, $\Psi$: Penalty coefficient, $\Phi$: Punishment coefficient, $\phi$: Step size factor, $x_{min}$: Lower bound of position, $x_{max}$: Upper bound of position, $round\left( x \right)$: The nearest integer number to $x$, $Eggs_{max}$: Maximum number of laying eggs, $Eggs_{i}$: Number of laying eggs for $Nest_{i}$, $Pa$: The possibility of the eggs being discovered by the host bird, $ELR$: Egg Laying Radius, $Iter_{max}$: Maximum number of iterations.