Understanding performance implications of nested file systems in a virtualized environment

Virtualization allows computing resources to be utilized much more efficiently than those in traditional systems, and it is a strong driving force behind commoditizing computing infrastructure for providing cloud services. Unfortunately, the multiple layers of abstraction that virtualization introduces also complicate the proper understanding, accurate measurement, and effective management of such an environment. In this paper, we focus on one particular layer: storage virtualization, which enables a host system to map a guest VM's file system to almost any storage media. A flat file in the host file system is commonly used for this purpose. However, as we will show, when one file system (guest) runs on top of another file system (host), their nested interactions can have unexpected and significant performance implications (as much as 67% degradation). From performing experiments on 42 different combinations of guest and host file systems, we give advice on how to and how not to nest file systems.