Parameters of load testing models: approaches to estimation

Abstract

The BigBlueButton video conferencing service was introduced to organize distance learning at BSUIR. Since April 2020, as Covid cases increased, the whole system load has increased drastically. During the testing phase, it turned out that increasing the capacity of the BigBlueButton server (BBB) relative to the nitial parameters did not significantly increase the performance of the system. There were some problems with the translation of images and connections to the system. The problems started at video conferences with more than 50 participants. Also, there were time synchronization errors, new user login errors, errors adding a video stream. It became necessary to combine several computing facilities (servers and server clones) into a single cluster. A load balancing service was implemented on the video conferencing system to distribute load flows. By combining multiple servers into a single information space, the service allows you to access more resources and distribute the load between the connected servers, depending on the number of users (Round Robin algorithm). After load testing of servers with different configuration in the cluster, the output parameters were obtained to analyze the performance of the cluster and the balancer.