Workload-based prediction of CPU temperature and usage for small-scale distributed systems

The recent boost in the usage of high-performance computing systems in small research environments, such as those found at many universities, stipulates the need of small-scale distributed systems. Owning to the rapid growth in both computing power and heat, development of proper thermal and resource management becomes crucial concern of the research community along with the vendors to ensure efficiency for such systems. Moreover, an accurate and relatively fast strategy is needed for adaptation of different sizes of workload in such systems. Therefore, in this paper, we focus on developing simple prediction models of CPU temperature and usage for the systems. We investigate impacts of macro-level parameters such as the number of machines and different sizes of workload on CPU temperature and usage via real experiment. Our experimental results reveal that for a certain size of workload, the variation in CPU temperature and usage is minimal in response to a change in the number of machines, which does not hold in the reverse way. Hence, we develop workload-based prediction models for CPU temperature and usage. We evaluate the accuracy of our models by comparing the values calculated based on these models against the measurements found from real implementation.