Applying of Quality of Experience to system optimisation
Applying of Quality of Experience to system optimisation
复制标题
将体验质量应用于系统优化
DOI：
--
复制
发表时间：
2013
期刊：
影响因子：
--
通讯作者：
B. Al
中科院分区：
文献类型：
--
作者：
Sascha Bischoff;Andreas Hansson;B. Al
最先进的移动的设备，例如智能手机和平板电脑，是高度通用的，并且必须跨多种应用提供高性能。感知到的性能和由此产生的用户体验可以成就或破坏设计。因此，设备设计和优化必须考虑最终用户的期望和感知，以及设备上运行的应用程序的类型和要求。传统上，设备优化侧重于低级别指标，如CPU浮点性能或GPU帧速率，而不是对最终用户最重要的方面。在这项工作中，我们调查的适用性体验质量（QoE）系统优化。我们定义了一组可测量的QoE指标，并围绕Web浏览器和两个图形基准运行一组实验。使用这些实验的结果，我们展示了使用QoE作为优化指标的优势，展示了在考虑用户体验的同时以最佳方式交换CPU性能以获得能源使用的能力。我们研究了两个GPU基准测试，以确定实现能效的理想内核数量，同时确保足够高的帧速率以保持高质量的用户体验。然后，我们将系统作为一个整体来看待，以及在不牺牲用户体验的情况下使用QoE优化整个系统的性能和功耗的可行性。我们实现了高达60%的系统能耗节约，对用户体验的影响有限。
State-of-the-art mobile devices, such as smart-phones and tablets, are highly versatile and must deliver high performance across a multitude of applications. The perceived performance and resulting user experience can make or break a design. It is therefore vital that device design and optimisation take into account the expectations and perceptions of the end user, as well as the types and requirements of the applications running on the device. Traditionally, device optimisation focuses on low-level metrics, such as CPU floating point performance or GPU frame rate, rather than on the aspects most important to the end user. In this work, we investigate the applicability of Quality of Experience (QoE) to system optimisation. We define a set of measurable QoE metrics, and run a set of experiments around a web browser and two graphics benchmarks. Using the results of these experiments, we show the advantages of using QoE as an optimisation metric by demonstrating the ability to optimally trade CPU performance for energy usage whilst taking into account the user experience. We investigate two GPU benchmarks to determine the ideal number of cores for energy efficiency, whilst ensuring a sufficiently high frame rate to maintain a high-quality user experience. We then look at the system as a whole, and the feasibility of using QoE to optimise performance and power consumption for the complete system, without sacrificing user experience. We achieve up to 60% savings in system energy usage with limited impact on the user experience.