Android/frameworks_base
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

AAH: Don't panic unless you are sure.

Browse Source 
Change the definition of the panic threshold for the time sync service
to take the round trip time of the packet producing the discipline
action into account.  The idea here is that, in the worst case, the
error in the measurement of the clock sync error might be as high as
the round trip time of the packet used to measure the error.  If a
packet says we are 50mSec fast, its RTT is 75mSec, then we could be
anywhere from 25mSec slow to 125mSec fast.

This change basically makes it so that the error measurement must be
greater than the RTT + the fixed panic threshold before we panic.  IOW
we don't panic unless we absolutely sure that our error must be
greater than the fixed panic threshold.  Also, lower the current value
of the panic threshold from 50mSec down to 20.  Now that the
definition takes into account the RTT, we can afford to be much
tighter in our definition of the fixed panic threshold.

Change-Id: Id801f6b59545ee52d59944dfa3d545d108f88b0e
This commit is contained in:
John Grossman
2011-10-13 14:02:52 -07:00
committed by Mike Lockwood
parent 9684be35bf
commit 89e058e36a
1 changed files with 12 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
services/aah_timesrv/clock_recovery.cpp
View File

@@ -183,9 +183,15 @@ bool ClockRecoveryLoop::pushDisciplineEvent(int64_t local_time,
filter_data_[min_rtt].point_used = true;
// Compute the error then clamp to the panic threshold. If we ever exceed
// this amt of error, its time to panic and reset the system.
// this amt of error, its time to panic and reset the system. Given that
// the error in the measurement of the error could be as high as the RTT of
// the data point, we don't actually panic until the implied error (delta)
// is greater than the absolute panic threashold plus the RTT. IOW - we
// don't panic until we are absoluely sure that our best case sync is worse
// than the absolute panic threshold.
int64_t effective_panic_thresh = panic_thresh_ + filter_data_[min_rtt].rtt;
delta = nominal_common_time - observed_common;
if ((delta > panic_thresh_) || (delta < -panic_thresh_)) {
if ((delta > effective_panic_thresh) || (delta < -effective_panic_thresh)) {
// PANIC!!!
//
// TODO(johngro) : need to report this to the upper levels of
@@ -233,8 +239,9 @@ bool ClockRecoveryLoop::pushDisciplineEvent(int64_t local_time,
applySlew();
}
LOGV("observed %lld nominal %lld delta = %5lld "
LOGV("rtt %lld observed %lld nominal %lld delta = %5lld "
"int = %7d correction %3d (P %3d, I %3d, D %3d)\n",
filter_data_[min_rtt].rtt,
observed_common,
nominal_common_time,
nominal_common_time - observed_common,
@@ -294,8 +301,8 @@ void ClockRecoveryLoop::computePIDParams() {
pid_params_.gain_I.doReverseTransform(pid_params_.correction_max, &tmp);
pid_params_.integrated_delta_max = static_cast<int32_t>(tmp);
// By default, panic when the sync error is > 50mSec;
panic_thresh_ = 50000;
// By default, panic when are certain that the sync error is > 20mSec;
panic_thresh_ = 20000;
}
void ClockRecoveryLoop::reset_l(bool position, bool frequency) {