Auto scaling is a service provided by the cloud service provider that allows provision of temporary resources to the subscriber’s systems to prevent overloading. So far, many methods of auto scaling have been proposed and applied. Among them, solutions based on low-level metrics are commonly used in industry systems. Resource statistics are the basis for detecting overloading situation and making additional resources in a timely manner. However, the effectiveness of these methods depends very much on the accuracy of the overload calculation from low-level metrics. Overloading is mentioned in solutions that usually favor a shortage of CPU resources. However, the demand for resources comes from the application running on that each application has the characteristics of demanding different resource types, with different CPU, memory, I/O ratios so it can not just be statistically on CPU consumption. The point of view here is that even though based on low level resources, the source for calculation and forecasting is the characteristic of the resource needs of the application. In this paper, we will develop an empirical model to assess the effect of the application's resource consumption characteristics on the efficiency of the lowmetricauto scaling solutions and propose an auto scaling solution that is calculated based on statistics of different types of resources. The results of the simulations show that the proposed solution based on multiple resources is more positive.