SAMPLING RATES
The sampling rate is defined as the number of times data is requested from a source by a client in a single second. A rate of 5Hz means 5 distinct samples in one second. Supermon uses peak sampling rate as a measure of how well it achieves LowPerturbation on the compute nodes and the interconnect pathways that lead to the client ultimately consuming the data.
Some sampling rates of interest are listed below:
- CPU level data is updated on a sub-second basis, usually on the order of 100Hz or higher.
Some sampling rates of interest are listed below:
- CPU level data is updated on a sub-second basis, usually on the order of 100Hz or higher.
- Network level data is updated on a sub-second basis, but slower than CPU information due to the logical and physical separation of network from CPU in the system. Expect 10Hz to give noticable changes in data over time.
- Hardware sensors are updated on a second-resolution basis. In some cases, buses such as the I2C bus, provide data at rates on the order of 1 to 1/5Hz. Sampling faster than this yields redundant and useless data.
Supermon provides, as part of the ProgrammingInterface, a set of functions that allow users to specify the set of data elements they want to sample along with their respective sampling rates. A schedule of samples is then created where each sample contains exactly the data that is desired, and the maximum amount of time is provided with no activity between samples to cut down on unnecessary, redundant samples that do nothing more than consume resources.
Figure 1: Two sampling schedules for different data sets. Schedule (a) requires one sample per second, while sample (b) eliminates the data with a high update rate showing how idle time can be introduced in the schedule to cut down on wasted cycles when the high update data is not required.