Udayan Bapat
2012-Feb-24 19:50 UTC
[libvirt-users] Regarding file vs block devices for the guest
Hi There is a general consensus in regards to use of a block device (exported partition) over a file device (.img or .qcow2) image. It is expected to see a big performance jump in case of a block device. However in my test I have seen it is not always the case. My testbed has Fedora-14.0 as the host and Ubuntu-10.0 as guest running on Intel x86 4 core/8 thread VTX enabled system. I ran file system benchmarks from the guest using iozone and measure throughput as well as CPU utilization. I observed that upto device sizes of 8 GB, file device (.img with ext3 partition) consistently outperformed block device (ext3 partition exported directly to the guest). Is similar anomaly seen by the community too? If not, i.e. block device should always outperform file, which layer should I be looking at to understand what could be happening. Just to make sure, I repeated same test a bunch of times to eliminate noise. As I increased the HDD size, then block device performance drastically improved and consistently performed better than fileIO. In both tests, my guest had 512 MB RAM with 1 vcpu. I ran all 11 iozone test to do performance comparisons. Any pointer/insight would be extremely helpful. Thank you in advance. -- Udayan -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://listman.redhat.com/archives/libvirt-users/attachments/20120224/eca80b63/attachment.htm>
Maybe Matching Threads
- [jra@samba.org: Patch to improve Samba write speeds on Linux ext3 with 3.2.x]
- Fully-Virtualized XEN domU not Booting over iSCSI
- a domain VTx with the VNIF does hang.
- more one question regarding gl and nouveau
- Re: PATCH 3/6 - direct-io: do not merge logically non-contiguous requests