human readable bitrates calculation

[junka]

Hi all,
I wonder if this is an old topic.

When I debug on a rate limit with sar -n DEV, there is always a deviation in the rxkB/s. I use the value to multiply 8.

When I set the limit to 40000kbps, the value in the sar output is 4883.12. Then I take it as 39065 kbps (4883.12 * 8).
At first , I doubted if there was any bug in my rate limit code. However, after days' digging, it turns out that the value from sar is not correct.

This value was calculated with the formula rxkb / 1024, so after calibration 39065 * 1024 /1000 = 40002, the value seems good enough.
So my question is why it is calculated in this way and should we fix it?

[junka]

seems duplicate, close.