Clouds of glass: fragility of virtual machine disk images

QEMU;qcow2;virtual machine;virtual hard disk;vulnerability;Redundant Arrays of Inexpensive Disks (RAID);block failure;corruption;disk image;fault injection;file system;checksum;convenience;robustness;reliability;fragility

Naha, Puskar ; Lumetta ; Steven S.

关键词: QEMU;    qcow2;    virtual machine;    virtual hard disk;    vulnerability;    Redundant Arrays of Inexpensive Disks (RAID);    block failure;    corruption;    disk image;    fault injection;    file system;    checksum;    convenience;    robustness;    reliability;    fragility;   

Others  :  https://www.ideals.illinois.edu/bitstream/handle/2142/50607/Puskar_Naha.pdf?sequence=1&isAllowed=y

美国|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

With virtual machines becoming common, it becomes important for them to be robust to errors. However, the virtual hard disk images used by virtual machines are not as reliable as physical disks. Existing solutions such as RAID can be used to improve reliability, but only at the cost of disk space usage and inconvenience.Using the virtual machine software QEMU, we show that its qcow2 virtual hard disk format is vulnerable to failures. We then introduce and analyze two alternative solutions to RAID1 (mirror) that use less disk space and are more convenient to use. As a baseline, the qcow2 format is the easiest to use, uses the least disk space, and has no runtime performance slowdown, but is vulnerable to corruption and block failures. Our first solution, qcow2r, protects qcow2 metadata from corruption and block failures, and it is designed to be only slightly more inconvenient to use. Also, qcow2r uses only 1.003 times the disk space and slows down run time performance by less than 1.5 times. Our second solution, qcow2c, protects all metadata and data in a qcow2 image from corruption and block failures. It is also designed to be as convenient to use as qcow2r. Using qcow2c requires 2.04 times the disk space and slows down run time performance by less than 2 times.In comparison, RAID1 is more inconvenient than our solutions, because it occupies at least half the provided disk space. It protects against block failures, but not corruption; thus, RAID1 is between qcow2r and qcow2c in terms of vulnerability. Also, RAID1 slows down run time performance by less than 1.5 times, which is comparable to qcow2r, but better than qcow2c. While our solutions target average users, they may be useful to enterprise cloud system operators as well.