Transaction-Aware Data Cluster Allocation Scheme for Qcow2-Based Virtual Disks
Minho Lee, Young Ik Eom · 2020
Recently, virtualization has become one of the backbone technologies for constructing data centers and cloud systems because it facilitates server consolidation through its attractive features. One major benefit of the virtualization technology is to support the physical server in concurrently running multiple VMs by logically partitioning the system resources into multiple ones. In constructing a virtualized system with multiple VMs, the qcow2 format is widely used as a virtual disk layout due to its storage space efficiency. However, the qcow2-based virtual disk has a critical drawback with respect to I/O performance. With the qcow2 format, excessive file sync operations are called in the process of data cluster allocation. In this paper, we introduce an efficient data cluster allocation scheme for the qcow2-based virtual disk to mitigate the performance overhead caused by the excessive file sync calls. For this, we adopt transactional support to the qcow2-based virtual disk and propose a new file sync operation, called gsync, which flushes the modified metadata in a bundle. Experimental results show that our scheme reduces the number of sync operations by up to 50.6% and improves I/O performance by up to 26.1%, compared with the conventional scheme.