| Trends show the popularity of solving the dual problems of over and under provisioning by replacing traditional server hardware with cloud services that can be quickly scaled with shifting traffic patterns. However, applications with a global customer base experience fluctuations in demand based on the location of users. Given a client-server architecture, where the server and database are deployed to a single location, there can be large disparities in response time, potentially leading to a loss of business in distant regions. Using Amazon Web Services, serverless architectures, and edge computing techniques, this paper tests 5 systems and a k-means analysis based data partitioning solution to this issue. The paper describes a methodology for adapting log data for use in the k-means algorithm, and compares distance measurements and centroid computations for use in the algorithm. Experimental results confirm that some of these approaches can significantly improve both response time and throughput. |
