I measure nodejs event loop using process.hrtime() also I count percentiles for measuring intervals. Here is my simple benchmark example
I have run this test on PC with CPU i7-4770.
This is how graph looks like when script runs on Windows 7(same type of graph is on OSX):
Asix X(horizontal) - time. Asix Y(vertical) - 90th percentile latency in μs (microseconds)
And this is how graph looks like when script runs on CentOS 7/6:
Asix X(horizontal) - time. Asix Y(vertical) - 90th percentile latency in μs (microseconds). Wavelength is about 2 minutes.
As you can see, in general latency on linux is bigger than on windows. But real problem is periodic peaks, when average latency and number of high latency are increasing.
I`m not sure is this OS, v8 or libuv issue. Is there any good explanation of this behavior? Or how can it be fixed?
