USAGE OF SESSION METRICS TO TASKS SCHEDULING IN BROWSER-BASED VOLUNTEER COMPUTING SYSTEMS

Abstract

Context. The study is devoted to the development an adaptive method for scheduling tasks in distributed browser-based volunteer computing systems. which is constructed on the session metrics of a volunteer user. This allows server to allocate tasks to each volunteer for computing, depending on the characteristics of his computing processes. More productive clients will receive more tasks, maximizing resource utilization, reducing server load and facilitating scaling; less productive clients will receive fewer tasks, solving the problem of excessive blocking of input data.Objective. The aim of the study is increasing the efficiency of distributed browser-based volunteer computing systems by using session metrics to tasks scheduling.Method. The session metrics that determine the behavior of volunteer users are described. The indicators that characterize the progress of the computational process on user systems were introduced. The formula to calculate the amount of input data for distribution to client is proposed. It takes into account the capacity of the client's system and the tendency of processes on its system. Simulation was done using different methods to calculate the amount of input data on the basis of the existing system of volunteer computing. To evaluate the performance of such systems, the total number of user hits to the server part and the total number of calculated results were used.Results. The proposed method is applied in the existing browser-based volunteer computing system. A comparison of the results of the proposed method with the existing ones showed a decrease in server load and an increase of the number of calculated results when session metrics are used.Conclusions. The simulation confirmed the convenience and efficiency of session metrics usage to tasks scheduling. The presented approach provides scalability of distributed browser-based volunteer computing systems. The prospects of further research are to use the users’ statistics and information from their browsers to calculate the amount of input data for distribution to client.