[Tinyos-devel] Ad hoc TimeSync group meeting during IPSN

[Omprakash Gnawali]

We had an ad hoc timesync group meeting during IPSN.

In attendance - Brano Kusy, Philipp Sommer, Răzvan Musăloiu-E., and
Omprakash Gnawali

Meeting notes:

We first discussed the challenges in multi-hop time sync. FTSP errors
can be large over multiple hops. People have tweaked the per-hop error
limit constant to get it to work reasonably over multiple-hops (for
example, Sukun's deployment on Golden Gate which had ~ 20 hops). Our
sense was it is not clear if GTSP solves this problem, it would
require further examination of protocols like GTSP. The general
guideline for FTSP is to use larger per-hop error tolerance rather
than the more intuitive smaller error tolerance per hop while using
FTSP over long hop networks.

Philipp, in his GTSP paper, shows high timesync errors with FTSP
within a neighborhood in which nodes are in direct radio
communication. Although GTSP has many nice properties, and obviates
explicit leader election, we concluded that we don't need all those
mechanisms to solve this particular problem. The problem is due to
fixed time propagation tree in FTSP. Adding jitters of a few seconds
to the periodic timer that drives FTSP beacons should introduce churn
to the FTSP time propagation tree solving this problem.

In many instances, we find that deployments have a collection root or
some roots. We want to be able to use those roots also as timesync
roots. This becomes handy when we have to synchronize PC's with the
motes. RATS is interesting work in this area. One concern is this
might increase the depth of timesync tree, which gets us back to
problem #1 above.

These are some directions in which time synchronization work might go
in the future. Comments welcome.

- om_p