View article

The focus of the sensor network community on energy-efficiency has produced a string of novel MAC, routing, and data-aggregation protocols. Their power consumption has mainly been assessed through simulations, ie by counting the fraction of time spent in sending, receiving, and computing, and multiplying that by figures taken from data sheets or isolated (single-node) power measurements. In contrast we present PowerBench, a 24-node testbed capable of recording the power consumption of all nodes in parallel with a 5kHz sampling rate and 30 µA resolution. This is accomplished by means of a low-cost interface board featuring a shunt resistor and an A/D converter, whose output is collectively sampled by an embedded Linux platform (Linksys NSLU2).

The experience with the PowerBench testbed so far is twofold. First, we have determined that–much to our surprise–timerbased estimations can match true, measured power consumption values within a few percent. Second, we have experienced that a graphical display of the power traces is an effective means to study (and debug) protocol behavior; in particular, inter-node related timing issues can be easily viewed from the state (IDLE/COMPUTE/RX/TX) changes embodied in the power data.