The performance parameters are parameters with which we measure and test the profitability and the performances of several static or dynamic load balancing algorithms.The static load balancing algorithms, share the client query between virtual machines in a data center for the processing. But, there is a problem when it comes to the current load of each virtual machine.The dynamic algorithms as "Efficient Response Time Load Balancer" and "Mini time processing load balancer" prove to be a solution to respond to this problem The advantage of these algorithms, before allocating a task, is that they search in the allocation tables on the virtual machine, with a metric which is quite inferior (be it the response time or the processing time).Tn this paper, we propose a new improvement of the load balancing by the algorithm « estimated finish time load balancer », that takes into account, the current load of the virtual machine of a data center and the estimation of the processing finish time of a task before any allocation, in order to overcome the problems caused by the static algorithms.The algorithm « estimated finish time load balancer» allows cloud service providers, to improve the performance, availability and maximize the use of virtual machines in their data centers.

Cloud users can have access to the service based on “pay as you go.” The daily increase of cloud users may decrease the performance, the availability and the profitability of the material and software resources used in cloud service. These challenges were solved by several load balancing algorithms between the virtual machines of the data centers. In order to determine a new load balancing improvement; this article's discussions will be divided into two research axes. The first, the pre-classification of tasks depending on whether their characteristics are accomplished or not (Notion of Levels). This new technique relies on the modeling of tasks classification based on an ascending order using techniques that calculate the worst-case execution time (WCET). The second, the authors choose distributed datacenters between quasi-similar virtual machines and the modeling of relationship between virtual machines using the pre-scheduling levels is included in the data center in terms of standard mathematical functions that controls this relationship. The key point of the improvement, is considering the current load of the virtual machine of a data center and the pre-estimation of the execution time of a task before any allocation. This contribution allows cloud service providers to improve the performance, availability and maximize the use of virtual machines workload in their data centers.

No abstract