Browsing by Subject "performance evaluation"

Energy usage and efficiency is an important topic in the area of cloud computing. It is estimated that around 10% of the world’s energy consumption goes towards the global ICT system [1]. One key aspect of the cloud is virtualization, which allows for the isolation and distribution of system resources through the use of virtual machines. In recent years, container technology, which allows for the virtualization of individual processes, has become a popular virtualization technique. However, there is limited research into the scalability of these containers from both an energy efficiency and system performance perspective. This thesis aims to investigate this issue through large-scale benchmarking experiments. Results of the benchmarking experiments indicate that not necessarily the total amount of containers, but the assigned task of each individual container are relevant when considering energy efficiency. Key findings show a link between latency measurements performed by individual containers and allocated CPU cores on the host machine, with additional CPU cores causing a drop in latency as the amount of containers increase. Further, power consumption seems to hit its peak when CPU utilisation is only at 50%, with additional CPU utilisation causing no increase in power consumption. Finally, RAM utilisation seems to scale linearly with the total amount of containers involved.