Disk Performance comparison for image File vs Raw File

[expertalert]

Good Morning.

We are trying to move to virtualbox heavly for all our servers.
but i need a help from virtualbox expert of disk performance report.

From Virtualbox website its saying :
"The actual performance difference for image files vs. raw disk varies greatly depending on the overhead of the host file system,"

So offcourse if i allow guest to use physical hardrive directly , it will improve the performane

but then problem is, I cant move those VM image from one Virtual box host to another virtual box host.
"Images which give access to individual partitions are specific to a particular host disk setup. You cannot transfer these images to another host; also, whenever the host partitioning changes, the image must be recreated."
the question is :

how much performance decrease if i use Virtual disk ?
is there any comparison between Virtualdisk V Raw disk ???

Please let me know
thanks
fosiul

[mpack]

The host OS implements a cache system for files which is bypassed if you use a raw disk. So, despite the myth, I would not expect raw disk to be faster.

The big overhead with a mechanical disk is the seek time. So, periodically defragment both host and guest (*) filesystems and you should be fine.

Don't use snapshots. Not so much because of the performance impact, but because compaction and defragmentation are nearly impossible if you use snapshots.

Note that a virtual drive has two levels of defragmentation to think about: virtual disk block order (a side effect of the order sectors were first written to) and the guest filesystem. Clone the VDI with CloneVDI to defragment the block order, then defragment the guest filesystem using the defrag tool of your choice inside the guest OS.

Don't believe anyone who tell you that Linux drives don't need to be defragmented. If they are used seriously then they most certainly do. Any filesystem which has a finite capacity, allows multiple files to be open at once, and/or allows a file to be extended sometime after it was first created - will eventually suffer from fragmentation if it's used "in anger" for long enough.

[expertalert]

Hi thanks

I am using Debian as Host.

I understand you using Windows as Host hence you are giving me advise for defrag .
As i use linux. I dont need to defrag the server as its not necessary .
Ref: http://geekblog.oneandoneis2.org/index. ... ragmenting
Ref: http://www.linux.com/archive/feed/32002

like you I belived that there should not be too much performance related issue if i use VirtualFixed disk.
but it would be really handly if virtualbox has any report to support this theory .

Thanks
Fosiul

[mpack]

The only reports I remember are bogus: those concerned made the beginners mistake of measuring performance with a virtual timer... and they of course get nonsensical results, most amusingly when they report the VM working at 10x the best speed the host is capable of.

I'm basing my comments on a reasonable understanding of how filesystems and mechanical drives work, as I prefer not to rely on anecdotal benchmark evidence. A benchmark tells you performance for a specific task at a particular instant in time (assuming you can believe the timing instrument and the competence of the tester). It doesn't tell you what the performance ought to be, or why it differs from some other approach.