[Tinyos-devel] nesC concurrency semantics, volatiles, etc.

[John Regehr]

> Generally, performance hits matter much more with respect to the running time
> of an interrupt handler than overall CPU utilization. Slowing down the FIFOP
> interrupt in the reception path, for example, might not add a significant
> amount of CPU overhead to an application overall, but it may cause a node to
> drop packets. Similarly, reducing the maximum ADC sampling rate from 8kHz to
> 2kHz is a big deal, although in both cases you'd (theoretically) be using the
> CPU close to 100%.

I hear you, but application-specific performance metrics are pretty 
difficult to deal with.  It would be nice if those of us doing tool work 
had a better sensornet performance metric!  Actually this would be more 
broadly useful, for example help help folks decide how fast to run the CPU 
or to decide between AVR, ARM, MSP430, etc.

John