Academia.eduAcademia.edu

Figure 12 – uploaded by Ahmadreza Montazerolghaem

See full PDFdownloadDownload figure
Figure 10. CPU consumption by the ELRC modules.

Figure 10 CPU consumption by the ELRC modules.

Related Figures (12)

Figure 1. Schematic of loMT in which a large volume of multimedia traffic of loMT applications is collected and sent to the loMT cloud servers through the network switches for storage, processing, and service.
Figure 1. Schematic of loMT in which a large volume of multimedia traffic of loMT applications is collected and sent to the loMT cloud servers through the network switches for storage, processing, and service.
Figure 2. a) Fixed loMT network capacity b) Variable loMT network capacity but unaware of demand (load) c) Variable loMT network capacity and aware of demand (NFV-based approach).
Figure 2. a) Fixed loMT network capacity b) Variable loMT network capacity but unaware of demand (load) c) Variable loMT network capacity and aware of demand (NFV-based approach).
In the following, we present the mathematical model for the optimal management of the IoOMT network resource management problem with constraints of resource, energy, and load.
In the following, we present the mathematical model for the optimal management of the IoOMT network resource management problem with constraints of resource, energy, and load.
Figure 3. Proposed Software-defined Internet of Multimedia Things framework.
Figure 3. Proposed Software-defined Internet of Multimedia Things framework.
Figure 4. ELRC controller modules (application layer). history of the CPU usages of PM1. For PM1, at is the value of the i-th CPU usage (0 < x} < 1). Moreover, y} is the value of the i-th memory usage and z} is the value of the the i-th hard disk usage (0 < y},z} < 1). Asa result, fi... (y) = Ui View ae) Yi_p+1l and fhara(2) = [zi 13 ee? Zi p+ are vectors that contain the history of the memory usages and hard disk usages of PM1. The goal is to obtain an estimate of Vit. (fi41), Yi+1 (Yi+1), and B+] (241) for each active PM, according to the f functions and a prediction method. The most important prediction methods include time-series analyses. Among these methods, the Normalized Least Mean Square (NLMS) predictor is the main one providing the best trade-off between complexity, accuracy, and responsiveness. That is why we apply this predictor as the prediction system in the Resource Load Estimation module. Next, we want to obtain @;.; for active PM k using the NLMS algorithm (#", , ). The ELRC has two parts: control and application. The SDN controller and NFV orchestration are responsible for controlling the OVSs and VNFs of the IoMT servers, re- spectively. They also issue the necessary rules based on the application layer modules. The three modules of Network Monitoring, Flow Monitoring, and Resource Monitoring explore and gather information using the Link Layer Discovery Pro- tocol (LLDP). The controller is predictive-driven to respond more quickly to the load changes. In this regard, the Re- source Load Estimation module estimates the future load of the network according to history. Also according to history, the IoMT Network Demarcation module determines the new size of the network. For this size, the IoOMT Server Selection module selects the best server, and the Path Selection module determines the appropriate path to reach the selected server. Finally, the decisions made are converted into appropriate rules by the Rules Handler module and delivered to the control layer. In the following, the details of ELRC modules are presented (Fig. 4).
Figure 4. ELRC controller modules (application layer). history of the CPU usages of PM1. For PM1, at is the value of the i-th CPU usage (0 < x} < 1). Moreover, y} is the value of the i-th memory usage and z} is the value of the the i-th hard disk usage (0 < y},z} < 1). Asa result, fi... (y) = Ui View ae) Yi_p+1l and fhara(2) = [zi 13 ee? Zi p+ are vectors that contain the history of the memory usages and hard disk usages of PM1. The goal is to obtain an estimate of Vit. (fi41), Yi+1 (Yi+1), and B+] (241) for each active PM, according to the f functions and a prediction method. The most important prediction methods include time-series analyses. Among these methods, the Normalized Least Mean Square (NLMS) predictor is the main one providing the best trade-off between complexity, accuracy, and responsiveness. That is why we apply this predictor as the prediction system in the Resource Load Estimation module. Next, we want to obtain @;.; for active PM k using the NLMS algorithm (#", , ). The ELRC has two parts: control and application. The SDN controller and NFV orchestration are responsible for controlling the OVSs and VNFs of the IoMT servers, re- spectively. They also issue the necessary rules based on the application layer modules. The three modules of Network Monitoring, Flow Monitoring, and Resource Monitoring explore and gather information using the Link Layer Discovery Pro- tocol (LLDP). The controller is predictive-driven to respond more quickly to the load changes. In this regard, the Re- source Load Estimation module estimates the future load of the network according to history. Also according to history, the IoMT Network Demarcation module determines the new size of the network. For this size, the IoOMT Server Selection module selects the best server, and the Path Selection module determines the appropriate path to reach the selected server. Finally, the decisions made are converted into appropriate rules by the Rules Handler module and delivered to the control layer. In the following, the details of ELRC modules are presented (Fig. 4).
Figure 5. a) Input membership function for BF, [0,1] b) Input membership function for Th [0,1] c) Input membership function for By [0,1] d) Output membership function for class* [-0.6,0.4].
Figure 5. a) Input membership function for BF, [0,1] b) Input membership function for Th [0,1] c) Input membership function for By [0,1] d) Output membership function for class* [-0.6,0.4].
Fig. 7 shows the topology and the testbed, including 8 PMs for the OMT VNFs, 2 PCs for the OVS switches and 1 PM for the ELRC controllers. PMs include the G9 (three), G8 (two), G7 (three), and G6 (one) HP servers. In the testbed, Floodlight v1.2, Open vSwitch v2.4.1, and Kaa server v1.2 are used to
Fig. 7 shows the topology and the testbed, including 8 PMs for the OMT VNFs, 2 PCs for the OVS switches and 1 PM for the ELRC controllers. PMs include the G9 (three), G8 (two), G7 (three), and G6 (one) HP servers. In the testbed, Floodlight v1.2, Open vSwitch v2.4.1, and Kaa server v1.2 are used to
Figure 7. Software-defined Internet of Multimedia Things testbed.
Figure 7. Software-defined Internet of Multimedia Things testbed.
(q) Number of VNFs & OVSs (scenario 1) (r) Number of VNFs & OVSs (scenario 2) (s) Number of VNFs & OVSs (scenario 3) (t) Number of VNFs & OVSs (scenario 4)
(q) Number of VNFs & OVSs (scenario 1) (r) Number of VNFs & OVSs (scenario 2) (s) Number of VNFs & OVSs (scenario 3) (t) Number of VNFs & OVSs (scenario 4)
Figure 8. Efficiency evaluation of data plane in various mechanisms by the QoS and QoE evaluation parameters.
Figure 8. Efficiency evaluation of data plane in various mechanisms by the QoS and QoE evaluation parameters.
Figure 9. Snapshot of PMs CPU consumption in different scenario.
Figure 9. Snapshot of PMs CPU consumption in different scenario.
cal scaling methods and compare them with the proposed method in this article (horizontal scaling). We will also study more sophisticated strategies to estimate the load of loMT servers. We also plan to develop the ELRC and the proposed framework instead of using VMs in the Docker-Container or OpenStack and compare them with each other. We will also look at using the Dual Decomposition or Lagrangian Relax- ation methods (instead of using heuristic ones) to reduce the complexity of the problem.
cal scaling methods and compare them with the proposed method in this article (horizontal scaling). We will also study more sophisticated strategies to estimate the load of loMT servers. We also plan to develop the ELRC and the proposed framework instead of using VMs in the Docker-Container or OpenStack and compare them with each other. We will also look at using the Dual Decomposition or Lagrangian Relax- ation methods (instead of using heuristic ones) to reduce the complexity of the problem.
Related topics:
Computer NetworksComputer Networking and TelecommunicationSoftware Defined Networks (SDN)Network Function VirtualizationIntenet of Things
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved