CPU Overheating Characterization in HPC Systems: A Case Study

Abstract

With the increase in size of supercomputers, also increases the number of abnormal events. Some of these events might lead to an application failure. Others might simply impact the system efficiency. CPU overheating is one such event that decreases the system efficiency: when a CPU overheats, it reduces its frequency. This paper studies the problem of CPU overheating in supercomputers. In a first part, we analyze data collected over one year on a supercomputer of the top500 list to understand under which conditions CPU overheating occurs. Our analysis show that overheating events are due to some specific applications. In a second part, we evaluate the impact of such overheating events on the performance of MPI applications. Using 6 representative HPC benchmarks, we show that for a majority of the applications, a frequency drop on one CPU impacts the execution time of distributed runs proportionally to the duration and to the extent of the frequency drop.

CPU overheating prediction in HPC systems

“…Concerning the running applications, even if we identified specific applications that generate CPU overheating events, only a few of their runs are impacted by an overheating event. More details about the analyses of these two points can be found in our previous publication 8 . We conclude from these results that the occurrence of overheating events does not have a single cause.…”

“…We study the occurrence of the CPU overheating events in a production supercomputer to identify their possible sources. The results presented in this paper are a summary of the main results that we described in a previous publication 8 . The analysis shows that there is no obvious correlation between the occurrence of overheating events and the load of the system, and that we cannot identify nodes that have a significantly higher probability of experiencing overheating events or applications that would increase the probability of overheating.…”

“…Before presenting our solution to predict CPU overheating events, we would like to present some experimental results that aim at assessing the impact of CPU overheating events on the performance of MPI applications that run over multiple nodes of an HPC system. Note that a more detailed description of these results are presented in our previous publication 8 …”

“…We also assess the impact of these overheating events on the running applications. Note that the results presented in this section are already presented in details in a previous publication 8 . Here, we only include a summary of the results that are useful for the design and evaluation of our predictive solution.…”

“…CPU overheating events are one of these events that can limit the performance of some jobs in supercomputers 8 . When the temperature of a CPU is too high, it significantly reduces its frequency to decrease its temperature and avoid hardware damage 9 .…”