[Tinyos-devel] Re: [Tinyos Core WG] Proposal for VariableSizedQueueC component

[Philip Levis]

On Sep 20, 2007, at 9:40 AM, Jan Hauer wrote:

>> A receive queue would
>> only help if the latency appears only between the RXFIFO.readDone and
>> its posted task.
>
> Yes, but this latency can be arbitrary.

Let me put this another way: if you have arbitrary task latencies,  
then the RXFIFO.readDone -> CC2420Receive.receive will not be your  
only worry; there are plenty of other places where it will occur,  
e.g., resource arbitration. This means that introducing a queue  
between the RXFIFO interrupt handler and the receive task will not  
solve the problem: you will still lose packets. Meanwhile, you're  
introducing an additional RAM requirement on the CC2420 stack.

If there were strong experimental evidence that the RXFIFO -> task  
latency is the dominant cause of packet drops, then David Moss (or  
application developers) could gauge the tradeoff between minimizing  
RAM utilization and improving packet delivery. But I think it's  
important to keep Amdahl's Law in mind.

Besides, if you have arbitrary task latencies, packet drops would be  
the least of your problems (e.g., timers).

Phil