VirtIO-trace: An Unified Tool for Analyzing I/O Characteristics on NVMe SSD in Virtualized Environments

Sewoog Kim, Jongmoo Choi

http://doi.org/10.5626/JOK.2018.45.4.332

Virtualization technology plays a significant-role in diverse computing environments such as cloud computing and data center. CPU manufacturers actively provide hardware-based virtualization techniques for virtualization systems. NIC manufacturers also support virtualization techniques to improve network I/O performance. However, there is a significant performance degradation in storage I/O virtualization, and many studies attempted to overcome this problem. Recently, NVMe(Non-Volatile Memory Express) SSDs(Solid State Drives) have become increasingly popular as storage devices for high-performance virtualized I/O systems. However, such fast storage devices cannot improve I/O performance significantly against one’s expectation. To optimize the storage I/O optimization performance, we need an I/O tracking and analysis tool. In this paper, we propose a novel tool that can monitor I/O behaviors on NVMe SSD in virtualized environments. The tool, which we refer to as VirtIO-trace, basically allows to trace I/O requests and their timing information like the existing blktrace. However, it differs from the traditional tools in that it provides NVMe SSD specific information such as queue status and submission/completion statistics, and virtualization specific information such as I/O processing time in VM/host systems. We implemented the tool in the KVM virtualization system. Experimental results show that the tool can collect I/O information in real time, which can be usefully exploited in analyzing I/O characteristics and exploring a new policy for enhancing performance and fairness on management of NVMe SSD in virtualization systems.