There has been an explosion of cloud services as organizations take advantage of their continuity, predictability, as well as quality of service and it raises the concern about latency, energy-efficiency, and security. This increase in demand requires new configurations of networks, products, and service operators. For this purpose, the software-defined network is an efficient technology that enables to support the future network functions along with the intelligent applications and packet optimization. This work analyzes the offline cloud scenario in which machines are efficiently deployed and scheduled for user processing requests. Performance is evaluated in terms of reducing bandwidth, task execution times and latencies, and increasing throughput. A minimum execution time algorithm is used to compute the completion time of all the available resources which are allocated to the virtual machine and lion optimization algorithm is applied to packets in a cloud environment. The proposed work is shown to improve the throughput and latency rate.
The best-known network-related problems can be solved using software-defined networks (SDNs) [
Cloud computing is pervasive and ubiquitous. Visualized applications are executed in the private or public cloud by one mobile user, and devices can be accessed throughout the world. Visualization is a core technology that dramatically increases the importance of network infrastructure. In distributed fashion, a combination of routers, switches, and other devices provides scale and reliability. However, in many scenarios, the network architecture cannot be used, which makes it difficult to satisfy requisites such as end-to-end quality of service load balancing and usage of VLANs [
SDN is an emerging network architecture that decouples network forwarding and control of functionality, as shown in
The introduction of SDN in cloud computing improves complexity and packet optimization issues and enables the hosting of millions of virtual networks without VLAN-like common separation isolation methods [
We provide a brief overview of the SDN-cloud network and discuss applications, objectives, and issues that motivate our work.
SDN is a programmable network architecture that decouples network control and forwarding functionality [ Cloud backbone enabled edge nodes connect the cloud provider and enterprise. Core nodes switch traffic between edge nodes [ An SDN-based controller enables configuration of forwarding tables in cloud backbone nodes and provides WAN network virtualization. Hybrid orchestration and cloud operation software manages provider and enterprise data center federation, resource management, and workflow, and facilitates multi-vendor networks between data centers, service providers, and users [
Customers can choose the best vendors and avoid vendor lock-in. DWDM, a PON-like technology, can be accessed with dynamic bandwidth. Timely migration and processing of inter-data center workload is possible. On-demand bandwidth is automated, and customer requirements are derived from intelligent service provisioning.
SDN can be used in data centers, the internet of things, cellular networks, and cloud computing to solve network-related issues.
Various issues are associated with packet optimization in the cloud with SDN.
Current cloud networking architectures rely on the paradigm of one size fits all. Application bandwidth requirements should be specified by cloud tenants for consistent performance. Guaranteed bandwidth is required for various tiered applications [ A wider variety of security appliances is needed, along with applications to accelerate tasks, cache, and balance loads. Access control and traffic isolation for end-users impact switch and router configurations [ Changes in requirements and various protocols can bring challenges to the operation and building of a cloud network [ For network-dependent failover mechanisms and IP addresses, out-of-box applications can necessitate reconfiguration and rewriting before deployment of the cloud.
Much work has been done in the fields of SDN and cloud computing.
Author | Approach | Purpose | Outcome | |
---|---|---|---|---|
Optimization in Cloud Computing | ||||
Rocha et al. [ |
Hybrid optimization model | Establishment of VM placement strategy for data centers | Balance between energy efficiency, network quality, and infrastructure | |
Liu et al. [ |
M/Geo/1 and M/Geo/m models based on queuing theory with embedded Markov chain and Z-transform | Optimize resource provisioning cost while satisfying customer QoS requirements | The proposed model can set up an affordable cloud firewall | |
Raad et al. [ |
Adaptive control framework | Improve access of distributed data center fabric delivered cloud services | Reduce transfer time by 80% to guarantee better user experience | |
Tian et al. [ |
Compressed TRIES algorithm | Phases of Hadoop and its improvement with the help of tuning | Solve problem of low efficiency | |
Indrawati et al. [ |
MINLP and Lingo software application | The model provides optimum solution that maximizes the revenue and provides cloud services as per the need and request. | Reduce cost and demand for internet and increase provider profit | |
Optimization in SDN | ||||
Qu et al. [ |
Optimization such as by grouping a short field in the search and merge phases and early termination | Large-scale packet classification on multicore processors | 33 million packets per second and 2000 ns per-packet latency with 49 of peak throughput for 32 K 15-field rule set by optimized decomposition-based approach | |
Yao et al. [ |
Particle swarm optimization (PSO) merging with maximum flow | Satisfying objectives of global network like maximum flow with forwarding table size limitation | A certain level of QoS fairness among flows, with better performance in data center and backbone network | |
Budhraja et al. [ |
Ant colony optimization technique and ML | Minimize data-transmission risks related to compliance and privacy | Preserve target compliance; use rules of intelligent eviction to obtain routing downloaded to switch | |
Aurizzi et al. [ |
Network function virtualization (NVF) and EHF | Counteract tropospheric fading | Use of SDN is good for intelligent management of proposed system | |
Raouf et al. [ |
ACOSDN | Optimize dynamic routing in SDNs | Handle dynamic network changes, reduce network congestion, and achieve high throughput with lower delay and packet loss rates | |
SDN-Cloud Integration Optimization | ||||
Amiri et al. [ |
Analytic hierarchy process-based game aware routing (AGAR), Dijkstra | Find optimum transmission path of a packet in a data center while maximizing bandwidth utilization and minimizing delay | AGAR was able to reduce end delay by 9.5 compared to Hedera routing, ECMP routing, and Dijkstra based on delay | |
Phan et al. [ |
M/M/k and M/M/1/∞ models | Proactive optimal resource allocation for elastic security VNFs in chaining service function | Defined mathematical requirements to analyze VNF resource allocation function and estimate number of packets in system | |
Tajiki et al. [ |
Exact solution and fast suboptimal one schemes | Estimate flow matrix | Simulation results show packet loss reduced by 20 compared to conventional approaches; throughput increased by 30 | |
Abdulqadder et al. [ |
PSO and chaotic secure hashing | User authentication and improved quality of service | OMNeT{+}{+} simulator implemented proposed Sec SDN-cloud and tested it in terms of bandwidth, latency, end-to-end delay, and packet loss, showing better performance |
The comparative study shows that optimization algorithms such as PSO, Chaotic, MINLP, and their combinations can improve results in terms of different parameters.
SDN enables the smooth integration of application provisioning in the cloud through automated interfaces, and fits naturally in cloud infrastructures such as IaaS, PaaS, and public or private clouds. SDN-based cloud networking solutions have advantages such as flexibility and scalability, but they support only certain network environments. The steps of the proposed approach are shown in
We have analyzed that in the offline cloud scenario, machines are deployed to process user requests where the applied queue environments evaluate the communication for efficient scheduling. Performance was evaluated to increase throughput and reduce latency, execution times, as well as bandwidth. During the first phase of implementation, minimum execution time algorithm has been applied to get the successful completion of allocated task, processes in the short interval of time without any error rate and Lion optimization has been used to reduce the bandwidth consumption and latency.
MET computes the completion time of the resources that are being allocated to the virtual machines.. The task is assigned to machines concentrated on the least execution time using essential facts of MET that sometimes deal with the great imbalance of the weight. The job is not conditional on machine accessibility.
Initially, the jobs are being provided to the machines from where we listed out the total quantity of jobs which are further handpicked on priority basis. The current development id and time are used to know the execution and completion time of the process respectively. Evaluate the performance time of the process by the machine. Evaluate the estimated time at which machine m is available to execute task t. The predictable or minimum completion time is estimated using the sum of the execution time to complete the processes by the machine and the initial execution time of the current task on the current processor or machine. Run this process for the completion of iterations and stop until all the jobs are completed. Figure out the minimum execution time of the task by the machine until all the jobs get exhausted.
The process of the lion algorithm is given below [ Initialize random populations Initialize prides and lions For each lion particle:
Select a random female lion for hunting Each female lion select the best position in the pride Weakest lion pride out from the population and become the nomad Each pride evaluate the imigration rate and become the nomad Evaluate the fitness function to select the best females and fill the mepty places which of the female lions which are migrated from the territory.
In
Step 1: | Assumptions for specifications of cloud computing: | |
Job count = 100; | ||
Total time = 50; | ||
Five queues xi (i = 1–5); | ||
Step 2: | Deployment of the 5 queues with their locations: | |
For i = 1:xi (i = 1–5) | ||
Plotting of machines specifications | ||
End | ||
Step 3: | Selecting Queue according to the priority which is the weighted round robin step | |
Step 4: | Apply MET algorithm: | |
Loop1 start: | ||
Loop2 start: | ||
Current_job=job_specification(3, i); | ||
current_time=job_specification(2, i) | ||
while total_time>0 | ||
arrival_time_job(i) = current_time; | ||
avail_time_mach(i) =(10–5).∗rand(1, 1) + 5; | ||
Estimated time the amount of time the resource r will take to execute the task | ||
estimated_time_mach = (time_per_job∗job_specification(4,:))/arr_time_job(i); | ||
ready_time = (avail_time_mach(i)∗job_specification(4,:))/arr_time_job(i); | ||
MET_val=est_time_mach + ready_time; | ||
minimum_computation_time(p) = min(MET_val); | ||
band_consumption(p) = sum(min_comp_time)∗job_count; | ||
total_time = total_time-1; | ||
end1 | ||
end2 | ||
Step 5: | Apply Lion optimization algorithm on result obtained by applying MET. | |
[rs, cs] = take the size of band consumption obtained from MET. | ||
iterations = rs; | ||
no_lions = cs; | ||
for 1:iterations | ||
normad_lions = band_consump(1:25); | ||
pride_lions =band_consump(25:50); | ||
hunters = numel(pride_lions); | ||
initial_nomad_pop = normad_lions(1:20); | ||
rest_nomad = normad_lions(numel(init_nomad_pop):end); | ||
Loop1: For i = 1: numel value of pride lions | ||
if i= numel value of normad lions | ||
break; | ||
end | ||
Find the best fitness and updated nomad | ||
end; | ||
end | ||
end | ||
Loop2: j = 1: numel value of updated nomad | ||
Get the best solution of lions | ||
End | ||
Step 6: | Use the final value obtained from Lion optimization for finding the different parameters. | |
Band consumption = min_comp_time∗(sum(best_sol_lions)/job_count); | ||
latency_met = (band_consump_met∗sum(min_comp_time))/job_count; | ||
throughput_met=((abs((band_consump_met))./(sum(min_comp_time)∗job_count))∗ 100); | ||
end |
The worst-case time complexity of the proposed algorithm is considered. This depends on the number of iterations, function evaluations, population size, and several loops. The ant lion's position initialization in a population of size NP has time complexity O(NP). Furthermore, the time complexity is used to calculate the cost values of the ant lions as O(NP)*O(F(x), where F(x) is the object/cost function. The complexity of the lion also depends on the iteration [
The due/packet-delivery dataset is used for the implementation. The dataset contains measurements of data delivery performance of a 802.15.4 link in an indoor scenario in which throughput, loss, delay, and energy were measured over six months under around 50,000 parameter configurations of seven key stack parameters, including around 200 million metadata transmission packets. There are two trace sets: delay and packet-metadata. The delay trace set has a pre-packet delay of packet transmission under various configurations of stack parameters. Metadata of each packet consist of noise floor, actual retransmission, arrival time, LQI, deliver success/fail, RSSI, actual queue size, and overflow or number traces.
We discuss below the generation of a cloud-based scenario using the MATLAB simulator with SDN, and how the proposed algorithms are used to optimize the performance for packet optimization.
To realize the proposed algorithm, a framework was developed and simulated in MATLAB.
Simulation time | Latency | Execution time (ms) | Total bandwidth (bps) | Throughput (Mbps) |
---|---|---|---|---|
10 | 4 | 15 | 3.4 | 0.75 |
20 | 3.5 | 5 | 2.1 | 0.6 |
30 | 3.8 | 27 | 2.9 | 0.46 |
40 | 2.5 | 18 | 3.1 | 0.54 |
50 | 3.7 | 22 | 3.2 | 0.65 |
Conventional networks are transformed into SDN by various enterprises to provide cost efficient and flexible network. But attacks and security breaches have exposed the weakness of SDNs. The most common, and most dangerous such attack is distributed denial of service (DDoS). Arivudainambi et al. [
SDN has become one of the most widely used network architectures to divide the control and forwarding planes. In the control plane, SDN centrally observes and regulates the network through software control, and manages the software defined by WAN, so there is a need for multiple controllers to handle the reliability and scalability issues of the network. To improve network performance, controller placement problem (CPP) is deployed which is threefold in nature i.e., the number of controllers to be placed in a network, the locations of these controllers and the assignment function of controllers to switches, in which all of them are important for the design of an efficient control plane. To get a reliable CPP, Verna-based optimization (VBO) was proposed by Ashutosh Kumar Singh et al. (2020) to reduce the total average latency of SDN [
Simulation time | Network B/W dependent DMMM algorithm latency (ms) | Network B/W dependent DMMM algorithm execution (ms) | LION optimization latency (ms) | LION optimization execution time (ms) | |
---|---|---|---|---|---|
10 | 50 | 0.88 | 4 | 15 | |
20 | 10 | 0.54 | 3.5 | 5 | |
30 | 15 | 0.33 | 3.8 | 27 | |
40 | 30 | 0.90 | 2.5 | 18 | |
50 | 25 | 0.64 | 3.7 | 22 |
Although the pride used in LOA is a stable social unit, it is a fission-fusion social group whose individuals are found in a range of subgroup compositions and sizes [
In this paper, specifications were taken as assumptions, such as the number of machines and allocated jobs per machine. Machines were deployed in MATLAB, and the minimum execution time along with LION optimization were used to solve the problem of packet optimization. Simulations were performed at five times with 10-second intervals, and results were obtained in terms of latency, execution time, bandwidth, and throughput. The results showed that the proposed method can solve the problem, and it gives better results than existing work. We generated a scenario in MATLAB that justifies the applicability of the proposed scheme in a real scenario. Although MET with LOA gave good results, it requires further testing. Premature convergence in LOA causes poor performance, and a strategy is needed that will not decrease the search space and will return to the original search space after a long period. The MET algorithm has limitations such as severe load imbalance. In the future, we will apply different hybrid metaheuristics and include the concepts of fog and edge computing. Testing is also needed on real-time datasets.