Uniform distribution of messages sent

[davidBar-On]

I think that one of the iperf3 issues that low bitrate is sent over network that supports (much) larger bandwidth, all the messages are sent at the beginning of the one second interval, and are not distributed over all the interval. It is desired in such case it will be possible to better simulate the case when the **traffic is evenly distributed in time**.

I would like to submit a PR to add that functionality, but I am not sure about the approach:

1. One approach can be to specify the period for sending messages, e.g. send a message every 10ms. This is similar to the approach suggested in PR #1075. (The random period can also be included.)
2. Another approach is to add an indication in the --bitrate option that sending the messages should be evenly distributed.

I am looking for comments about which approach is better and/or suggestions for other possible approaches.

Note that one of the main issues for both approaches is that in many cases the resolution of the OS timers is not high. E.g. a resolution of 10ms is quite common. This issue can be mitigated as done in PR #1075, by estimating the minimum resolution, use a period of that resolution and send burst of messages each period. E.g. if a message should be sent every 1ms but the estimated timer resolution is 10ms, then a burst of 10 messages will be sent every 10ms.

UPDATE

This discussion is irrelevant and is my mistake. iperf3 messages sending is already evenly distributed and the messages are not sent in bursts at the beginning of each sec. This is done by iperf_check_throttle() by setting green_light based on the total average bitrate from the beginning of the test, and not per the total bits sent in the last sec as I thought.

PR #1343 was submitted with a proposed option of random gap delay between packets.