A Holistic Evaluation of Docker Containers for Interfering Microservices
A Holistic Evaluation of Docker Containers for Interfering Microservices
复制标题
Docker 容器干扰微服务的整体评估
DOI：
--
复制
发表时间：
2018
期刊：
影响因子：
--
通讯作者：
R. Ranjan
中科院分区：
文献类型：
--
作者：
D. N. Jha;S. Garg;P. Jayaraman;R. Buyya;Zheng Li;R. Ranjan
容器技术（例如 Docker、LXC 等）的进步通过提供虚拟机管理程序的轻量级替代方案改变了虚拟化概念。 Docker 已成为最流行的容器管理工具。最近关于容器与虚拟机管理程序和裸机的比较的研究表明，容器在几乎所有情况下都可以实现裸机性能。然而，当前的文献缺乏对实验评估的深入研究，以了解托管在单个或跨多个容器中的微服务之间的性能干扰。在本文中，我们提出了同时运行异构微服务集的 Docker 容器性能评估的实验研究。我们按照 CEEM（云评估实验方法）进行了一系列全面的实验，以测量运行竞争或独立微服务的容器之间的干扰。我们还考虑了通过显式指定 cgroup 来限制容器资源的影响。我们根据容器间（由两个并发执行的容器引起）和容器内（在容器内执行的两个微服务之间引起）干扰来评估容器的性能，这在当前文献中几乎被忽略。评估结果可用于对容器化环境下微服务智能资源供应的干扰效果进行建模。
Advancement of container technology (e.g. Docker, LXC, etc.) transformed the virtualization concept by providing a lightweight alternative to hypervisors. Docker has emerged as the most popular container management tool. Recent research regarding the comparison of container with hypervisor and bare-metal demonstrates that the container can accomplish bare-metal performance in almost all case. However, the current literature lacks an in-depth study on the experimental evaluation for understanding the performance interference between microservices that are hosted within a single or across multiple containers. In this paper, we have presented the experimental study on the performance evaluation of Docker containers running heterogeneous set of microservices concurrently. We have conducted a comprehensive set of experiments following CEEM (Cloud Evaluation Experiment Methodology) to measure the interference between containers running either competing or independent microservices. We have also considered the effects of constraining the resources of a container by explicitly specifying the cgroups. We have evaluated the performance of containers in terms of inter-container (caused by two concurrent executing containers) and intra-container (caused between two microservices executing inside a container) interference which is almost neglected in the current literature. The evaluation results can be utilized to model the interference effect for smart resource provisioning of microservices in the containerized environment.