BLAFS: A Bloat Aware File System

CoRR(2023)

引用 0|浏览20
暂无评分
摘要
While there has been exponential improvements in hardware performance over the years, software performance has lagged behind. The performance-gap is caused by software inefficiencies, many of which are caused by software bloat. Software bloat occurs due to the ever increasing, mostly unused, features and dependencies in a software. Bloat exists in all layers of software, from the operating system, to the application, resulting in computing resource wastage. The problem is exacerbated in both cloud and edge setting as the number of applications running increase. To remove software bloat, multiple debloating tools have been proposed in the literature. However, these tools do not provide safety guarantees on the debloated software, with some files needed during run-time removed. In this paper, We introduce BLAFS, a BLoat-Aware-file system for containers. BLAFS guarantees debloating safety for both cloud and edge systems. BLAFS is implemented on top of the Overlay file-system, allowing for file-system layer sharing across the containers. We compare BLAFS to two state-of-the-art debloating tools (Cimplifier and Dockerslim), and two state-of-the-art lazy-loading container snap-shotters for edge systems (Starlight and eStargz). Our evaluation of real-world containers shows BLAFS reduces container sizes by up to 97% of the original size, while maintaining the safety of the containers when other debloating tools fail. We also evaluate BLAFS's performance in edge settings. It can reduce the container provisioning time by up to 90% providing comparable bandwidth reductions to lazy-loading snap-shotters, while removing 97% of the vulnerabilities, and up to 97% less space on the edge.
更多
查看译文
关键词
blafs,file
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要