A raspberry pi and a cloud server were set up to share the workload between the two. I used face recognition algorithm as an example of a heavy workload for the raspberry pi, in order to find out if it is better to share the workload to the server for processing the image. The image was taken from the pi camera and sent to the server to let the server process, so there are 2 bottlenecks in this situation, the network bottleneck and the processing speed of raspberry pi. The network bottleneck was determined by sending image taken from the pi camera to the server through ftp protocol in different resolutions and with different network bandwidth, the process bottleneck was determined by changing different image resolution and process on both platforms. I used the data obtained from measuring the time it takes to send different image resolutions to determine the bottleneck of network and data obtained from the time it take to process an image on both platforms to determine the bottleneck of process speed of raspberry pi in image processing, I also compared the data obtained from the experiments to determine at which point of network bottleneck it is better to process on raspberry pi and vice versa.
My results showed that it is better to process the image on the raspberry pi if we have a slow network(under 1Mbps) in all resolutions due to the overhead of ftp protocol, and it is better to send to the server when we have fast internet(over 1Mbps) because the processing speed of the server used in this project is more than 2 times faster than the raspberry.
|Uddannelser||Datalogi, (Bachelor/kandidatuddannelse) Kandidat|
|Udgivelsesdato||18 dec. 2017|