With the exception of patch #1, all patches in this changeset
are concerned with RTT estimation.
Patch #1: Avoids the accumulation of large send credit. This patch
has been discussed for a while. It is not yet clear whether
the upper bound should be one t_ipi or one RTT. This patch
uses t_ipi as a safe and conservative bound. This could be
changed later, but without a bound it will clearly happen
that large send credits will build up which will cause bursty
traffic over a disproportionally long time.
Patch #2: Sin of omission fixed - we didn't use a fallback RTT as per RFC 4340.
Patch #3: Stops CCID3 sending trivial amounts of elapsed time. I found this out
while reverting the interface timestamp patches submitted in April.
The point is that with the window counter CCVal as RTT indicator and
the CCID3-specific feedback mechanism, the elapsed time will amount
to tiny amounts of time between calling the send routine, since feedback
is sent immediately: this amounts to elapsed time in the order of < 50 usec.
Patch #4: The interface of dccp_sample_rtt is relieved of some of its burden.
Patch #5: More accurate detection of idle times (I think this comes closer to CCID3 now).
Patch #6: Inline for moving average - thanks to Arnaldo.
-
To unsubscribe from this list: send the line "unsubscribe dccp" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
