Broker Algorithm Optimizes Cloud VM Scheduling Through SLA Negotiation

Cloud computing platforms deliver services to customers over the internet, and one of the most crucial steps in that environment is virtual machine scheduling, which determines how physical machine resources are located for launching a virtual machine instance. A new open access study published in Discover Networks on 9 August 2026 proposes a broker-based service level agreement negotiation framework that integrates an optimization algorithm called Hybrid Leader-Based Optimization, or HLBO, to schedule virtual machines in cloud computing. The research, authored by S. Supreeth and Kiran Kumari Patil, reports that the framework achieved a utility of 0.7193, a negotiation time of 0.6648, and a request satisfaction of 0.9245 when evaluated with 10 virtual machines over 10 negotiation rounds.

Service level agreements are significant strategies for guaranteeing the performance of cloud services, and they improve trust between cloud service providers and cloud service customers. The SLA negotiator works to lessen the gap and offer confidence between the cloud service supplier and the consumer, approving consumers to succeed in the important service efficiently after constructing the best probable decisions for negotiation with numerous service providers. Requests submitted by users may require different quality of service requirements in the form of an SLA, comprising minimum queue time, maximum response time, and satisfying the job limit constraint. SLAs summarize essential controls for guaranteeing quality of service, fault-tolerant capability, cost, security, and service validity.

In the devised model, virtual machine scheduling is carried out among cloud service users, user agents, brokers, and the cloud service provider. Each cloud service user contains multiple user agents, and each user agent approaches multiple brokers for searching the optimal provider. The broker agent acts as an intermediary, or middleware, between service users and service providers, and is responsible for searching the optimal cloud service provider for virtual machine allocation. The broker performs this search using the HLBO algorithm by considering fitness parameters such as utility, negotiation time, and request satisfaction. After the optimal provider is recommended to the user agent, the service level agreement for the requested service is monitored. If the SLA between the cloud service user and the provider is acceptable, resources are interchanged directly; otherwise, mediation of the SLA negotiation is completed through the broker-based framework.

The HLBO algorithm is a population-dependent optimization method that mimics the population updating process of certain individuals. Each population member represents a candidate solution, and in this research each solution encodes the index of a cloud service provider. The algorithm updates positions based on three dissimilar individuals: one random member, the best member, and the corresponding member. A hybrid leader is generated for every population member at every iteration, with participation coefficients computed from the quality of each candidate solution relative to the worst objective function value. During the exploration phase, the hybrid leader guides members across various locations of the search space, while the exploitation phase permits a local scan for better solutions near the attained solutions, with a step size that shrinks as the current iteration approaches the maximum number of iterations. New positions are accepted only when the determined new objective function is superior to the old location, and the process continues until the best solution is achieved.

The selection of an optimal cloud service provider is formulated as a multi-objective optimization problem. The broker agent aims to simultaneously maximize negotiation utility and request satisfaction while minimizing negotiation time. Because HLBO requires a single objective value for evaluating candidate solutions, the multiple objectives are combined using a weighted aggregation approach, with the overall fitness expressed as F = αJ + βQ − γT, where J represents utility, Q indicates request satisfaction, and T denotes negotiation time. Equal importance was assigned to all three measures, so the weight coefficients were set to α = β = γ = 1/3, satisfying the constraint that the weights sum to one. Negotiation time was normalized to the range from 0 to 1 using min–max normalization prior to fitness computation to ensure compatibility with the other metrics. The authors state that the weights were not individually tuned through a separate optimization procedure, and that a separate sensitivity analysis was not performed because the objective was balanced consideration of all three performance measures.

The framework was implemented using the Python programming environment integrated with the CloudSim simulator, with experiments conducted on a system running Windows 10 with an Intel Core i3 processor. The simulation environment was configured to model a cloud infrastructure consisting of multiple data centers, hosts, virtual machines, and user workloads. Hosts were characterized by a processing capacity of 10,000 MIPS, 16 GB RAM, 1 TB storage, and 10 Gbps bandwidth, while virtual machines were configured with varying capacities ranging from 1000 to 3000 MIPS, 2 to 8 GB RAM, and 100 GB storage to simulate heterogeneous cloud resources. A time-shared virtual machine scheduling policy was adopted to allocate processing resources among virtual machines, and the scheduling process was executed under dynamic workload conditions with synthetic workloads representing multiple cloud service users submitting requests with varying processing requirements, execution times, and SLA constraints.

Comparative analysis was conducted against four existing approaches: broker-based SLA negotiation, agent-based dynamic SLA negotiation, a multi-objective optimization method, and a power-aware virtual machine selection policy. These baselines were implemented within the same CloudSim simulation environment to ensure a fair and consistent comparison, with settings established according to the descriptions in their original studies and identical workload conditions, virtual machine configurations, and simulation settings maintained across all methods. With 10 virtual machines and 10 rounds, the utility of the broker-based approach was 0.5757, the agent-based dynamic approach 0.5614, the multi-objective optimization method 0.5855, and the power-aware selection policy 0.6532, whereas the HLBO-based model reached 0.7193. Negotiation time for the proposed model was 0.6648, compared with 0.9644, 0.8637, 0.7640, and 0.7339 for the four baselines respectively. Request satisfaction reached 0.9245 for the proposed model against 0.8687, 0.9067, 0.8990, and 0.8832 for the baselines, corresponding to performance improvements of 6.03 percent, 1.92 percent, 2.75 percent, and 4.46 percent.

Scalability experiments were also conducted with 15, 20, 25, and 30 virtual machines. With 15 virtual machines, the utility of the proposed model was 0.7996, compared with 0.5857, 0.5887, 0.5943, and 0.6285 for the four baseline methods, and request satisfaction reached 0.9339 against 0.8730, 0.8777, 0.8783, and 0.8927 for the baselines, yielding improvements of 6.51 percent, 6.01 percent, 5.95 percent, and 4.40 percent. According to the reported scalability analysis, as the number of virtual machines increases the utility value gradually improves, the negotiation time exhibits only minor variations across configurations, and request satisfaction remains satisfactory across all virtual machine sizes. The authors interpret these findings as demonstrating that the framework can scale with increasing cloud infrastructure size while maintaining reliable performance.

Statistical validation was performed using 30 independent simulation runs for both the 10 and 15 virtual machine configurations. For each competing method, the mean and standard deviation were computed for utility, negotiation time, and request satisfaction, and Welch’s t-test was employed to assess the statistical significance of performance differences, with a significance level of p < 0.05. According to the study, all pairwise comparisons between HLBO and the baseline approaches produced p-values below 0.05, indicating that the observed improvements are statistically significant and not caused by random simulation variation. The authors attribute the improved performance to the optimization capability of the HLBO algorithm in the broker selection process, noting that when the number of virtual machines is moderate the search space becomes more balanced, enabling the strategy to effectively explore and exploit candidate solutions.

The study acknowledges limitations. A direct convergence comparison with other optimization algorithms such as genetic algorithm, particle swarm optimization, and ant colony optimization was beyond the scope of the present study and is left for future work. The authors also note that the devised method failed to establish a security function in the virtual machine scheduling, and they propose that future enhancements could include security functions to preserve against unauthorized persons entering the network, thereby enhancing trust in the SLA negotiation model. Additionally, because the broker agent is selected by a single optimization algorithm, the authors suggest that future work could combine two or three optimization methods into a hybrid algorithm to further improve the model.

Subject of Research: Broker-based SLA negotiation using the Hybrid Leader-Based Optimization algorithm for scheduling virtual machines in cloud computing

Article Title: Broker-based SLA negotiations with optimization for scheduling virtual machines in cloud computing

Article References: Supreeth, S., & Patil, K. K. (2026). Broker-based SLA negotiations with optimization for scheduling virtual machines in cloud computing. Discover Networks, 2(1), Article 27. https://doi.org/10.1007/s44354-026-00040-3

Image Credits: AI Generated

DOI: 10.1007/s44354-026-00040-3

Keywords: cloud computing, virtual machine scheduling, service level agreement, SLA negotiation, Hybrid Leader-Based Optimization, broker agent, cloud service provider, CloudSim, multi-objective optimization, request satisfaction, negotiation time, utility