How important is high-rate data handling?

One of the main selling points of mCerebrum is its high-rate data handling. Impressive improvements over other architectures such as the AWARE framework are presented. However, it is less clear to me how likely there is a need for such a high throughput.

Concretely, which studies that currently rely on the mCerebrum framework have the highest data samples collected per second? How many samples, and would running these studies not have been possible without the described optimizations?

In order to better interpret ‘number of samples’ it would also be useful to have some overview of concrete sensors and the amount of data samples they contribute per second. The SenSys paper states one sample is a double and 300Hz is listed for accelerometer, gyro, magnetometer, gps, light, microphone, and barometer. How should this be interpreted exactly? There is some room for ambiguity here (e.g., for x, y, z independently, or the whole ACM measure as a whole?)

Most, if not all, of the studies we deploy mCerebrum for, have data rates that exceed 350Hz of continuous data collection from a variety of devices. These field deployments typically range from 14-30 days and result in a minimum of 423 million data points up to a maximum 907 million points per individual.

The paper only breaks down the aggregate sampling rate by device due to space constraints. Since this has not been published yet, I do not want to discuss it in much detail here at this time. The paper presents an analysis of total data rate capabilities and you can work out the effects of how the data representation affects effective sampling rate on a per sensor basis.