Data collected from the Physical Stress application onto small computing devices. Telemetry collected from the sensorization of a RaspberryPi v.4B performance, serving TorchServer queries, under environmental stress, such as temperature, humidity and voltage changes, in a controlled test chamber. The device has been stressed with ambient temperatures up to 105ºC and humidity from 8% to 82%, also voltage changes from 4.6V to 5.2V, while queries with variable VideoAnalytics load from an external machine. Ambient metrics have been taken from the controlled chamber sensorization (out-of-board sensors), form the same device (resources usage from OS), and from the client side (quality of service).

ApacheBench, Version 2.3, Revision 1903618

Python, 3.12.3

TorchServe, 0.12.0

Environmental physical set-up for source experiments: * Controlled environment in a closed cardboard chamber, containing the Raspberry Pi v.4B device, and containing the DH11 sensors for humidity and temperature, with orifice for cables. The Arduino data collector is kept outside of the chamber. The chamber is heated with a "mug heater" from 55ºC to 75ºC to achieve ~85ºC inside the chamber and ~105ºC inside the RaspberriPi CPU. Also, the chamber is humidified with a purified water diffusion humidifier to increase humidity from 8% to 82%, though an opening in the cover, avoiding directing the water diffusion to the board or sensors. The RaspberryPi is powered by a programmable power source, applying voltages 4.6V to 5.2V, as minimum and maximum according to the manufacturer indications, also from users reporting device failures outside such range. Logical stress is performed from an external machine with higher computational power, connected to the RaspberryPi via ethernet cable, varying the number of concurrent queries to stress the device CPU, while collecting the sensors data from the Arduino via USB. * The telemetry and sensors data have been recorded raw from sensors, client and operating system, to reproduce a real environment with potentially variation on sensors, controllers and client data reporting. INTEL UFunding #14780