This fixes a subtle bug in the calculation of the inter-packet gap and shows
that t_delta, as it is currently used, is not needed. And hence replaced.
The algorithm from RFC 3448, 4.6 below continually computes a send time t_nom,
which is initialised with the current time t_now; t_gran = 1E6 / HZ specifies
the scheduling granularity, s the packet size, and X the sending rate:
t_distance = t_nom - t_now; // in microseconds
t_delta = min(t_ipi, t_gran) / 2; // `delta' parameter in microseconds
if (t_distance >= t_delta) {
reschedule after (t_distance / 1000) milliseconds;
} else {
t_ipi = s / X; // inter-packet interval in usec
t_nom += t_ipi; // compute the next send time
send packet now;
}
1) Description of the bug
-------------------------
Rescheduling requires a conversion into milliseconds, due to this call chain:
* ccid3_hc_tx_send_packet() returns a timeout in milliseconds,
* this value is converted by msecs_to_jiffies() in dccp_write_xmit(),
* and finally used as jiffy-expires-value for sk_reset_timer().
The highest jiffy resolution with HZ=1000 is 1 millisecond, so using a higher
granularity does not make much sense here.
As a consequence, values of t_distance < 1000 are truncated to 0. This issue
has so far been resolved by using instead
if (t_distance >= t_delta + 1000)
reschedule after (t_distance / 1000) milliseconds;
The bug is in artificially inflating t_delta to t_delta' = t_delta + 1000. This
is unnecessarily large, a more adequate value is t_delta' = max(t_delta, 1000).
2) Consequences of using the corrected t_delta'
-----------------------------------------------
Since t_delta <= t_gran/2 = 10^6/(2*HZ), we have t_delta <= 1000 as long as
HZ >= 500. This means that t_delta' = max(1000, t_delta) is constant at 1000.
On the other hand, when using a coarse HZ value of HZ < 500, we have three
sub-cases that can all be reduced to using another constant of t_gran/2.
(a) The first case arises when t_ipi > t_gran. Here t_delta' is the constant
t_delta' = max(1000, t_gran/2) = t_gran/2.
(b) If t_ipi <= 2000 < t_gran = 10^6/HZ usec, then t_delta = t_ipi/2 <= 1000,
so that t_delta' = max(1000, t_delta) = 1000 < t_gran/2.
(c) If 2000 < t_ipi <= t_gran, we have t_delta' = max(t_delta, 1000) = t_ipi/2.
In the second and third cases we have delay values less than t_gran/2, which is
in the order of less than or equal to half a jiffy.
How these are treated depends on how fractions of a jiffy are handled: they
are either always rounded down to 0, or always rounded up to 1 jiffy (assuming
non-zero values). In both cases the error is on average in the order of 50%.
Thus we are not increasing the error when in the second/third case we replace
a value less than t_gran/2 with 0, by setting t_delta' to the constant t_gran/2.
3) Summary
----------
Fixing (1) and considering (2), the patch replaces t_delta with a constant,
whose value depends on CONFIG_HZ, changing the above algorithm to:
if (t_distance >= t_delta')
reschedule after (t_distance / 1000) milliseconds;
where t_delta' = 10^6/(2*HZ) if HZ < 500, and t_delta' = 1000 otherwise.
Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
de6f2b59e5