Video compression for interactive cloud gaming

This master thesis aims to develop an implementation of JPEG 2000 that works on the cloud efficiently. The development starts from an open-source implementation of JPEG 2000, that is OpenJPEG. Among the most important features of cloud computing are flexibility and on-demand scalability, which in technical terms means, to be able to exploit the facilities on the cloud, the workload should be distributable on different nodes, which reduces the processing time significantly depending on the number of nodes the user decides to spawn. Moreover, this includes the ability to increase the processing power on-demand. The new multi-threaded version of OpenJPEG is more compatible with the environment of the cloud and takes better advantage of the facilities it provides. The approach started first by browsing the available options of a solution implementation on the cloud. This first step decided which approach to take and which changes are to be applied to the software (OpenJPEG). The second step was to develop a benchmark tool to asses the performance of OpenJPEG and to compare it to the performance of another implementation of JPEG 2000, that is Kakadu. This way, we have a typical product to compare to. After that, the development of a new solution was carried out and the results were obtained. To be able to efficiently optimize OpenJPEG on the cloud, a profiling of its performance was carried out in order to identify the bottlenecks of the performance, then compromising those bottlenecks was a starting point for optimizing OpenJPEG, since the best boost that can be obtained in performance, can be realized by reducing bottlenecks. In this thesis, time performance is the target and is the number-one-priority of optimization to be done. The bottlenecks of time consumption in OpenJPEG were in encoding codeblocks in the compression operation and decoding codeblocks in the decompression operation. As obtained from the results of the benchmarking tool, improving these bottlenecks resulted in an improvement in time consumption but it increased memory usage which is normal due to multi-threading. The thesis also included testing the resulting optimized software and presenting the results.