Re: [Xen-users] Understanding sparse-files

[John Haxby <john.haxby@xxxxxxxxxx>]

Rustedt, Florian wrote:
> What exactly is the advantage of sparse-files against "normal" files
> with fixed length?
>
>   
There are both advantages and disadvantages.

> First i thought this is something like an auto-increasing file. But if i
> take a 2GB partition and add two sparse-files with 1GB each, i can't add
> an additional one, the disk is full?
>
>   
No, that's not it.

> So what about this mystic advantage? Is it only the faster creation of
> that file with dd, because it is not completely filled?
> That's all?
>   
If you create yourself a nice big sparse file like this

   dd bs=1M seek=10240 count=0 if=/dev/zero of=huge

And then look at what you've got with "ls -lh" you'll see you have a 10G 
file that was created almost instantly.  On the other hand, "ls -sh" 
will show that the file is actually occupying no space at all (well, 
almost no space).  You can make this file bigger like this:
   dd bs=1M seek=20480 count=0 if=/dev/zero of=huge

and this will make it 20GB and still not occupying much space.

I suspect you already know this, but if you didn't, you do now :-)

The advantage of this 20GB file is precisely that it occupies next to no 
space on the disk that holds it.  I can start writing data into it (that 
is, use it a a guest's disk) and the blocks needed will be allocated as 
they are used.  In fact, I could have a 200GB guest disk image even 
though the disk I have at the moment is only 120GB and I'm using quite a 
lot of it -- it would only be a problem if the guest actually wanted to 
use all that space.
There are some problems with sparse files: the compress beautfully (gzip 
reports 99.9%) but it takes a while to read the empty space and when you 
uncompress the file you discover that it now actually occupies disk 
space: there's no good way to distinguish between an unallocated block 
and a block full of zeroes.   This also means that you need to be 
careful how you back these files up: you need something a little 
cleverer than gzip.
Another problem with sparse files, especially when using them as domU 
disks is that blocks that are contiguous in the file are not contiguous 
on the disk.   That means if, in the guest, if you just "dd if=/dev/xvda 
of=/dev/null" then domU will be seeking back and forth all over the 
place to return the blocks in the order that they're being asked for.   
You don't need xen for this --  when I downloaded the DVD image of 
Fedora 10 using transmission (a bittorrent client) a checksum on the 
resulting file only managed to read it at about 4MB/s.  On the other 
hand, when I copied the file the checksum on the copy ran at closer to 
100MB/s -- bittorrent clients like transmission really ought to 
pre-allocate the disk space to that you get something contiguous and 
also not embarrassingly run out of space half way through.
In a nutshell, though:

pros: over-committed disk space

cons: performance

jch



_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users