Closed elishafer closed 1 year ago
Hi @elishafer,
I do not see a noticeable difference between the two graphs. The data was obtained by 5 minutes of benchmark and I would assume that any visible difference would be caused by the small sample size (normally you would run these tests for several hours/days) as well as background load. As Docker is based on abstractions on the Linux kernel level , it should achieve native performance on a Linux machine. After all a Docker container still uses the same kernel as the host system! The only aspects that might suffer a slight performance hit are those where some form of virtualization is involved, such as I/O and networking. Latter can be worked around by removing the isolation between the container's network stack and the host systems with net=host
. For actual benchmarks you might have a look at this fairly old research paper from IBM (also available for free on IBM Research).
Thanks for the quick and detailed response! Also thank you for your guides, your motivation section for academy really struck a chord with me.
Glad to hear that! Let me know if we can close this issue.
In the Example section you have two graphs. One with docker and one without. It seems that the test without docker has a higher latency threshold than the one with docker. Do you perhaps know the reason why this happens? I would think that a system with docker would have a higher latency threshold.