Clouds execute very heterogeneous applications such as webservers and video streaming that have different characteristics. For example, one application burns mostly CPU whereas another uses network heavily. Some applications vary their characteristics...
Clouds execute very heterogeneous applications such as webservers and video streaming that have different characteristics. For example, one application burns mostly CPU whereas another uses network heavily. Some applications vary their characteristics with time. Yet, resource scheduling in clouds focuses on a single resource -- CPU scheduling or network scheduling. We call it a separate scheduling. Through extensive performance evaluation and profiling, this paper points out the limitations of the separate scheduling, which results in the performance degradation of the time-varying application. Evaluation results show that the separate scheduling only offers 67% and 73% of maximum performance on average. Based on the evaluation results, we argue that it is necessary to develop a new scheduling technique that can overcome the limitation of the separate scheduling and support cloud applications with time-varying characteristics.