Abstract
The success of the technology of cloud computing lies mainly in its business model of pay-as-you-go where users pay only for the resources really consumed. However it is known that there is no warranty for the QoS that these resources will provide at runtime. In this paper, we suggest an approach to make load balancing more dynamic to better manage the QoS of multi-instance applications in the Clouds. This approach is based on limiting the number of requests that, at a given time, can be effectively sent and stored in queues of virtual machines through a load balancer equipped with a queue for incoming user requests. This limitation is intended on the one hand to allow requests to go on to the faster instances, and on the other hand to better mitigate the effects of interference of sharing resources by the fact that a large part of the requests which were intended to instances that have become affected by degradation are still stored at the load balancer and can be allocated to non-affected instances or to new instances which will be created. A performance study using the simulator CloudSim showed the gain that this approach can generate, compared to classical approaches of load balancing.