[Tinyos-devel] problem of using FTSP and CTP

[Miklos Maroti]

I believe that you can do it, but please do it in a platform
independent way, so it works both on the CC2420 and RF230. The whole
point is that we do not want to reimplement time sync every time
someone wants to use a single-hop time sync message, and this is why
we implemented TimeSyncActiveMessageC. Miklos

On Mon, Mar 23, 2009 at 7:29 PM, Andreas Köpke <koepke at tkn.tu-berlin.de> wrote:
> Yep, I do propose to implement FTSP on top of an AMSenderC.
>
> On Montag 23 März 2009 Miklos Maroti wrote:
>> Hi Andreas,
>>
>> On Mon, Mar 23, 2009 at 7:09 PM, Andreas Köpke <koepke at tkn.tu-berlin.de>
> wrote:
>> > I agree with Phil here: components that use the AMSend virtualizer should
>> > not be influenced by components that do not use it.
>> >
>> > But is it really necessary for FTSP do use ActiveMessageC directly? I
>> > think not. In my opinion, there are two reasons why FTSP uses
>> > ActiveMessageC directly, one is ease-of-porting and the other, more
>> > important one, is that it uses a message _footer_ on tmote. On the sender
>> > side, the footer contains the time when the message was generated,
>> > whereas on the receiver side it contains the time that passed between the
>> > message generation time and the time its SFD was transmitted. A footer is
>> > a sensible choice here, since it allows the CPU enough time to compute
>> > the deltaT and modify the message on the fly.
>> >
>> > So the fact that FTSP attaches directly to ActiveMessageC is actually a
>> > hack that circumvents the lack of dynamic footers in message_t. As  a
>> > case study, I implemented a time synchronization on top of the AMsend
>> > virtualizer, using a standard AMSenderC component. It works, I just have
>> > to handle the footer the way TimeSyncActiveMessageC does: attach a
>> > uint32_t at the end of the message, set it to the current local time and
>> > expect that it has been modified to contain deltaT on reception. Perfect,
>> > simple and without any side effects on other components that use AMSend.
>> > Unfortunately, it requires that the application is "footer aware" and
>> > "time stamp aware" -- no big deal for FTSP, though.
>>
>> FTSP does not use ActiveMessageC (look it up), it uses the
>> TimeSyncActiveMessageC and that is not a hack. Of course
>> TimeSyncActiveMessageC could be implemented on top of the AMSenderC
>> and not the ActiveMessageC. Maybe that is what you are proposing? That
>> would work. By the way, we are not talking about FTSP or CTP per se,
>> that is just where people see the bad interaction of
>> TimeSyncActiveMessageC, AMSenderC and ActiveMessageC.
>>
>> Miklos
>