云计算逐渐成为一种主流的基础服务，容器由于其轻便型是目前云计算之中最常使用的一种虚拟化抽象。基于镜像这样的模板容器可以迅速部署应用在云服务器上。然而现有的容器引擎在大规模镜像分发时容器会产生冷启动时间过长的问题。本文分析并验证了影响Docker容器冷启动的关键因素，提出了一种大规模容器镜像分发加速模型。通过以文件为粒度的延迟加载，以及按使用顺序的文件分层，加快了镜像的传输速度，进而加速容器启动。基于该模型，实现了镜像分发系统D4C (Doing Deft Distribution of Docker Container)。并从容器冷启动时间、启动镜像大小、网络传输量三个方面进行测试，验证了D4C在这几个方面的优势。 Cloud computing has gradually become a mainstream basic service, and containers are currently the most commonly used virtualization abstraction in cloud computing due to their lightness. Based on such template containers, applications can be quickly deployed on cloud servers. However, the existing container engine will have a long cold start time when the container is expanded and distributed. The key factors affecting the cold start of Docker containers are analyzed and verified here, and an expanded container stack acceleration model is proposed. By taking the file as the model, the distributed system D4C (Deft distribution to Docker containers) is realized. Tests were conducted in terms of container cold start time, boot image size, and network transmission volume to verify the advantages of D4C in these aspects.
展开▼
机译：云计算逐渐成为一种主流的基础服务，容器由于其轻便型是目前云计算之中最常使用的一种虚拟化抽象。基于镜像这样的模板容器可以迅速部署应用在云服务器上。然而现有的容器引擎在大规模镜像分发时容器会产生冷启动时间过长的问题。本文分析并验证了影响Docker容器冷启动的关键因素，提出了一种大规模容器镜像分发加速模型。通过以文件为粒度的延迟加载，以及按使用顺序的文件分层，加快了镜像的传输速度，进而加速容器启动。基于该模型，实现了镜像分发系统D4C (Doing Deft Distribution of Docker Container)。并从容器冷启动时间、启动镜像大小、网络传输量三个方面进行测试，验证了D4C在这几个方面的优势。 Cloud computing has gradually become a mainstream basic service, and containers are currently the most commonly used virtualization abstraction in cloud computing due to their lightness. Based on such template containers, applications can be quickly deployed on cloud servers. However, the existing container engine will have a long cold start time when the container is expanded and distributed. The key factors affecting the cold start of Docker containers are analyzed and verified here, and an expanded container stack acceleration model is proposed. By taking the file as the model, the distributed system D4C (Deft distribution to Docker containers) is realized. Tests were conducted in terms of container cold start time, boot image size, and network transmission volume to verify the advantages of D4C in these aspects.
展开▼
