**6. Related works**

Messina [12] proposes an agent-based model to provide System Level Agreement negotiation. The authors use an ontology to establish the resources needed for each submission. Ontologies are used to provide knowledge to the environment. Therefore, it is possible to establish a more adjusted allocation of resources according to the application's input parameters. Ontologies generate knowledge about the semantic rules to perform the correspondence between the requested resource and the available resource. The work proposed in [12] uses an ontology to provide knowledge, however this approach does not handle situations not foreseen by the ontology well. The context-based strategy of the model proposed in this article provides the best configuration for running applications based on the updated state of the system. The approach proposed in this article performs the allocation of resources considering, in addition to the performance of the application, the performance of the system, which does not happen in the work proposed by [12].

Das uses a resource scheduling policy based on [13] artificial intelligence. However, Das uses the Teaching-Learning Based Optimization learning algorithm as a basis for his proposal. The justification given by the authors for the adoption of the learning algorithm in the context of resource allocation in computational grids is that Teaching-Learning Based Optimization is considered a light and efficient algorithm to find the global solution to optimization problems. Another work based on artificial intelligence is presented in [14]. The authors use an approach centered on estimating values for various data transmission parameters, such as latency and use of *links*. The approach used in [14] is applied only to provide service guarantees, based on QoS prediction through fuzzy logic. Parameters, or formulations, for controlling overall performance are not specified.

In [15] a dynamic resource allocation method, is proposed for load balancing in fog environments. For this, the method has a scheduler capable of performing dynamic migration of services to achieve load balancing for computing systems in fog. Negative aspects related to migration, QoS degradation, and QoE, are not considered in the [19] work. In the works [16, 17] scheduling strategies with real-time constraints are discussed. In [16], aspects of QoE and QoS are addressed but not in depth in order to propose directives to measure and improve these attributes. In [17], the authors develop a work in order to investigate how utility is affected by performance parameters in environments focused on fog aimed at healthcare applications. To evaluate the use of a fog data center, the resources of the iFogSim tool were used. In the work [18], a new resource allocation algorithm based on stable correspondence is proposed, in order to benefit users and providers in the fog environment. However, the authors do not clearly show how the aspects involving user satisfaction and the performance of the environment are treated. **Table 5** shows that our proposal establishes a model that meets QoE and makespan together. This tradeoff is not achieved in the works analyzed so far being our main contribution.

In [19] is proposed a pricing policy based on the QoE. This QoE is expressed from result of the allocation and try to optimize resource allocation from statistical information of the computational requests. This mentioned strategy is implementable in real-time brokers according to the authors. optimal dynamic allocation rule based on


#### **Table 5.**

*Comparison between proposals.*

the The developed solution is statistically optimal, dynamic, and implementable in real-time. The proposal in [20] is based on an energy and collaborative model in load balancing. The proposal in [19] is based on a statistical and dynamic model aiming users QoE. In [20] is proposed an algorithm to both: meeting the application latency requirements and providing energy efficiency in the heterogeneous edge tier. To the algorithm proposed in [20] implements a collaboration strategy at the edge of the network aiming at heterogeneous environment characteristics. According [20] is possible to reduces the waiting time for meeting requests. The proposal in [20] is based on an energy and collaborative model in load balancing. **Table 5** summarizes related work and shows a comparison with our proposal in terms of QoE, environment performance (makespan), and technical approaches to implementation.
