Overlit: New Storage Driver for Localization and Specialization
Inhyeok Kim, Kwon Je Oh, Young Ik Eom · 2019
The container technology, which is an operating system level virtualization technology, has been widely used in server computing environments such as cloud systems and data center systems, to overcome the limitations of conventional hypervisor-based virtualization technologies. Docker, which is developed primarily for Linux, aims to support high performance and convenient management of the server systems, while sacrificing some level of isolation. It supports very powerful storage and volume drivers to split a single image of each application into several sub-images for the repeatability and deduplication, but it still has some problems, such as global failures and bundled performance, caused by sharing a single host filesystem among all the layers. We develop a new container storage driver for localization and specialization, where the former is for eliminating deep dependencies among the layers and the latter is for enabling each layer to use its own filesystem. Our experimental results show that our storage driver improves the performance of read-only layer by up to 15% and that of read-write layer by up to 5 times over the other drivers.