The benchmarking consortium EEMBC has released the first of series package designed to measure the energy efficiency of IoT wireless communications. IoTMark-BLE analyzes the way in which different hardware platforms handle Bluetooth communication for a wide range of scenarios.
Markus Levy, president of EEMBC, said the Bluetooth benchmark will be followed by versions that measure the energy performance of Thread, LoRa, and Zigbee, among others. "There are many different connectivity options for the IoT, and they all have an energy cost."
One of the first issues the IoT-Connect working group that developed the Bluetooth benchmark was the proliferation of profiles. "We asked, which profile do we use? Healthcare? Home automation?" Levy asked.
The working group took the option of building a framework to measure the different tasks needed to relay packets over a wireless connection using different parameters. With that data, it is possible to estimate performance for the many different use-cases possible in IoT applications.
The working group put together a set of hardware tools based on off-the-shelf boards such as the Raspberry Pi. These interface with the device under test to perform the power measurements. Users who want to perform their own benchmarks can buy the platform as a kit or assemble their own using the EEMBC software.
A typical measurement is one for transmission-intensive applications. The MCU being tested reads data from a serial channel, performs a filtering operation then notifies the radio transceiver, adds a checksum and returns to sleep. This is repeated ten times with a sleeping period of approximately one second in between each run. Manufacturers will be able to provide a single score that reflects the overall efficiency of their platform. "But the idea is to be able to break it down, and see the actual events and figure out where to make optimizations," Levy said.
Brent Wilson, co-chair of the IoT-Connect working group and director of applications engineering at Silicon Laboratories, said vendors and users will have the freedom to offload processing to hardware – they will not have to run the entire protocol on MCU if the offload option is more energy efficient.
"EEMBC benchmarks like CoreMark were highly prescriptive. There were certain workloads that could not be touched. But these IoT workloads we describe at the operational level. If you have a programmable engine that can schedule these different tasks without requiring intervention from the CPU, that's entirely legal," Wilson said. "The benchmark verifies that it's completed correctly because it looks at the data. It's to encourage the hardware vendors to show their strengths."