Server implementations for improving the performance of CAR systems based on mobile phones

Collaborative Augmented Reality (CAR) systems allow multiple users to share a real world environment, including computer-generated images in real time. The hardware features of most current mobile phones include wireless network capabilities that offer a natural platform for CAR systems. However, the potential number of clients in CAR systems based on mobile phones is much larger than on CAR systems based on other kind of mobile devices, requiring a system design that takes into account scalability issues. This paper presents the experimental characterization of CAR systems based on mobile phones, providing quantitative results about well-known performance metrics in distributed systems. The results show that the system saturation point depends on the overall percentage of CPU utilization in the computer platform acting as the system server. Also, the results show that it is limited by the server I/O in some cases. Therefore, this paper also shows different server implementations that can improve the system throughput, as well as a comparative study. The results show that the implementation based on UDP messages provides a significant improvement in system throughput with respect to other implementations based on TCP, at the cost of losing a very small percentage of messages. These results validate this implementation as the most efficient one for large scale CAR applications based on mobile phones.