[Tinyos-help] send - sendDone delay (MICAz, T2)?

[Philip Levis]

On Jan 3, 2008, at 5:58 AM, Murray, Ben wrote:

> Hi
>
> I did some tests a while ago with a couple of MICAz nodes when  
> putting some
> time sync into my code and came across a 10ms randomness in the delay
> between my code calling .Send() and the .sendDone() event being  
> signalled.
> The delay seemed to vary pretty randomly between 3ms and 13ms. Is  
> there some
> reason for this (large?) delay? I based the timing on the  
> difference in
> value of a millisecond timer for matched send and senddone pairings  
> and used
> that to achieve decent enough (hopefully about +-1ms) sync for some
> small-scale controlled tests but it was a bit of a rushed (and rather
> roundabout) way to achieve accurate sync and not really practical in a
> proper deployment.
>

It's because of the MAC layer. The default CC2420 layer in T2 has an  
initial backoff of approximately 1-10ms. So after you call send(),  
the MAC layer waits 1-10ms before trying to send the packet.


> I saw the discussion on time sync in the devel forum and people  
> seem to
> indicate rather better sync in the order of (micro seconds) us or  
> better and
> I was wondering where the option to stick a timestamp in the outgoing
> messages at the actual time of transmission rather than the time of  
> calling
> the send event as the latter is not very accurate requiring a more  
> complex
> sync proceedure? the TEP I read indicated that the timestamp in the
> message_t is not transmitted but is local to the node?

The core WG is talking about this right now. The actual embedment of  
a timestamp is not a big deal; the question is what the API should  
be. We're also trying to wrestle with network interoperability  
questions. Take a look at Miklos' mail to -devel in November. It has  
two interfaces; one is for receiver-only timestamps, the other is for  
transmitter-receiver timestamps.

Phil