Nowadays, the technics of health applications that use the cloud are being developed. However, the existing methods are static and cannot approve dynamic changes in the dynamic environment (for example, when the network and virtual machines (VMs) have a change in resources values) during the execution process. Since the cloud environment provides virtualized resources for computing and for storage (for example: health information) with used many virtual machines. Also, data applications require communication between these virtual nodes, placement of VMs and data location to achieve overall computation time. The majority of scientific researchers present in the current literature that the selection of physical nodes to place data and virtual machines as not separate problems. In addition, in the cloud environment, the major challenge is network security. So, there is no better solution than firewalls which are used to filter packets (detect spam packets). But the problem is that the cloud has a dynamic topology, for that these firewalls cannot examine the content from inside the packet and the network becomes vulnerable. This traditional firewall they only provide basic protection at the network layers and cannot work in complex topologies like Cloud Computing. For this, the risk that we will have unprotected areas. Regarding this challenge, in this article, we proposed to divide the cloud topology into zones, so that each zone is supervised by a controller. Thus, each virtual machine is supervised by a firewall. For remote network saturation with exchanged data between controllers and VMs, the number of controllers must be minimized and the addition of a new VM must be well placed in our new architecture (Divided-Cloud). For this, in this work, we used a learning method of Machine Learning (ML) "Decision Tree" at the level of the addition of a new controller. According to the affected result, the algorithm reaches its maximum accuracy which is equal to 83%. Furthermore, about the location of a new VM, we used a “KNeighborsClassifier” calcification method and it gives an accuracy is equal to 83%.