admin管理员组

文章数量:1278952

We have designed benchmarks to test execution times of some algorithms coded within a library that we are working on.

Those algorithms are mono-threaded. So the algorithms can use at most 100% of the CPU power, i.e., a single core (as displayed from the command top).

We have usually other applications running on our machine. It is not easy to stop them all each time we want to launch the benchmarks (and some of them are launched at start).

Our computers are multi-core Xeon computers. Our Linux distribution is Ubuntu.

When we launched those benchmarks, we noticed a high variability of about 20% in the recorded times for the same workload running on the same computer (the typical time of a benchmark is ~ 50 seconds). It is probably due to competition with other programs running on the machine. This variability occurs even when we take care to not ignite CPU consuming tasks when the benchmarks run and it makes it quite difficult to assess the evolution in the algorithms' efficiency.

Is there a way to attribute CPU/RAM resources with higher priority to our benchmarks, so that it would be less perturbed by other programs?

I tried to use the nice command, but it did not seem to have any effect. I have seen that there are a tools named TuneD that might be useful. Do you think it is something to look at more deeply?

本文标签：

版权声明：本文标题：ubuntu - Tackling variability when recording execution times while benchmarking a library under Linux - Stack Overflow 内容由网友自发贡献，该文观点仅代表作者本人， 转载请联系作者并注明出处：http://www.betaflare.com/web/1741252865a2366119.html， 本站仅提供信息存储空间服务，不拥有所有权，不承担相关法律责任。如发现本站有涉嫌抄袭侵权/违法违规的内容，一经查实，本站将立刻删除。