PXI tackles processor benchmarks
By Richard A. Quinnell, Contributing Technical Editor -- Test & Measurement World, 6/1/2006
Predicting the performance of an embedded processor in an application is a challenging task that requires careful control and measurement. The Embedded Microprocessor Benchmark Consortium (EEMBC) has developed a series of software tests to facilitate making such predictions as well as to facilitate comparison-shopping among design alternatives (Ref. 1). Now, EEMBC is using PXI technology to tackle its latest benchmark effort: power utilization.
 |
| The EEMBC EnergyBench platform measures the current and voltage drawn by the device under test. Courtesy of National Instruments. |
A key factor in most portable system designs is the power consumed during operation. This power usage affects battery size, operating lifetime between battery charges or changes, packaging design for thermal management, and weight, all trading off against device performance and features. The ability to evaluate the power performance of design alternatives is thus a significant need in the portable device market.
The EEMBC has addressed this need with the creation of its EnergyBench power and energy benchmark. EEMBC president Marcus Levy explains that EnergyBench simultaneously measures system performance and power consumption of embedded processors with all of their built-in I/O peripherals. The resulting score is a metric that product developers can use to evaluate their designs and that processor vendors can use to promote their product's efficient use of energy.
To make benchmark scores useful for industry-wide comparison, however, EEMBC has to certify the scores. "For us to certify benchmark scores," said Levy, "we need vendors to send in their test platforms. We don't want all kinds of homemade versions being sent in, so we defined a standard platform." That standard platform uses PXI instruments running LabView software to control the test and make measurements.
Control of timing is keyAs shown in the figure, the benchmark platform monitors the power source voltage and measures the current passing to the device under test (DUT) by using a sense resistor. This requires only a modest data-acquisition capability that is not unique to PXI. The control and timing features of PXI, however, are critical to making the right measurements.
According to Levy, the platform starts and stops the running of benchmark software and keeps the execution synchronized with the power measurements. "We need to coordinate data collection with the benchmark run," said Levy. "Further, we needed to control the timing of the sampling. The scores are based on several runs of the benchmark software, and we need to alias the sampling times on the multiple runs to ensure that we are not sampling on the same processor event each time."
To fully exercise the processor and I/O, the platform also generates signals and measures response times for such parameters as I/O and interrupt latency and context switch times. The platform further controls overall test timing. Because power consumption varies with temperature, for instance, the platform exercises the processor for 30 min to warm it up before running the benchmark. The platform also automatically completes the required number of runs.
The choice of PXI as the basis for the power benchmark is due partly to its ability to provide all of the necessary timing and control functions. Cost was another issue. "EEMBC customers are always concerned with their budgets," said Levy. Finally, there was the human element. "We considered making a custom board, but that becomes a political nightmare as we address the question of which processor to use," Levy explained. "PXI was a much easier approach politically as well as technically."
The PXI standard platform for processor power benchmark testing is only the beginning of PXI's use within EEMBC. The same test setup can be used to evaluate telecom and networking devices using the appropriate EEMBC benchmark software. PXI's modularity also allows it to accommodate new benchmarks as they become defined. "We could add new hardware if needed or even just make changes at the LabView level," said Levy. Automotive benchmarks based on the PXI platform are already planned.
| REFERENCE |
|
