BUILDING ACCURATE MODELS TO DETERMINE THE CURRENT CPU UTILIZATION OF A HOST WITHIN A VIRTUAL MACHINE ALLOCATED ON IT

摘要

In cloud computing environments there are several virtual machines running in the same host. This fact opens the door for possible side channel-attacks. Prior to perform an attack it is mandatory to determine co-residency with the victim. An synchronized variation in the CPU activity in the host is a possible indicator of the presence of neighboring processes. However, cloud providers do not give information about the hosts CPU load, so we have to figure out a way of estimating it. We estimate the host CPU load considering its impact on the performance of a virtual machine (VM) running on it. In this work, we show that it is possible to calculate the CPU load of the host by executing a reference process, measuring the time it takes to execute, and using this information as an input to generate the CPU load models. We explore regression methods and regression methods tuned with genetic algorithms for the model generation. As a result, considering a CPU load value between 0 (no load) and 100 (maximum load), we obtain models which compute the host load with a mean squared error of around 5%, 10% and 30% (depending on the host architecture) when estimating the load every second.