Characterizing OS Behaviors of Datacenter and Big Data Workloads
Zheng Chen, Jianfeng Zhan, Zhen Jia, Lixin Zhang · 2016
As datacenters and big data workloads become dominant ones, the pressure of new system design achieving cost-effectiveness rises, for both architecture and operating system communities. Consistent efforts on benchmarks have been taken to characterize the micro-architectural characteristics of those workloads. Statistics show that datacenter and big data workloads suffer from more front-end pipeline stalls than those of traditional workloads. In simulator-based architecture researches, the operating system (OS) execution is often ignored. However, for real-world applications, the OS execution plays an very important role. As an attempt to shed some lights on the evolutionary approaches for improving OS execution on data center and Big Data workloads, we quantitatively break down the OS behaviors that contribute to pipeline front end stalls. We run workloads from BigDataBench, and to identify OS behaviors that contribute to these stalls. The insights derived from evaluation help us identify key limits of current OS architectures. Our studies on OS behaviors also naturally lead to several OS evolutionary recommendations to efficiently manage the diversity of scale-out data center and big data workloads. We further show that, with appropriate changes in the OS kernel, the overall application performance can be improved due to the reduction of front end stalls.