qemu-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Qemu-devel] [PATCH] Add new block driver for the VDI format (use ai


From: Jamie Lokier
Subject: Re: [Qemu-devel] [PATCH] Add new block driver for the VDI format (use aio)
Date: Tue, 28 Jul 2009 09:34:52 +0100
User-agent: Mutt/1.5.13 (2006-08-11)

Amit Shah wrote:
> On (Mon) Jul 27 2009 [10:23:19], Jamie Lokier wrote:
> > Kevin Wolf wrote:
> > > Though I guess that for static images typically not only metadata is
> > > preallocated, but zeros are written for the whole disk content? Maybe we
> > > could implement a three-way flag like preallocate=[no,metadata,data] and
> > > let qemu-img handle the data part (writing zeros is the same for all
> > > formats and would even work with raw).
> > 
> > Note that you can also preallocate space with posix_fallocate(), which
> > fills the file with zeros but (sometimes) doesn't take as long as
> > writing zeros.
> 
> It won't take as long as writing zeroes if the filesystem underneath has
> support for fallocate(). ext4, btrfs, xfs have support for fallocate().
> 
> > Apparently it is almost essential when writing large files in small
> > pieces on Windows, and on Linux it is supported by the ext4
> > filesystem, but I haven't checked either claim.
> 
> I did some comparisons:
> 
> http://log.amitshah.net/2009/03/comparison-of-file-systems-and-speeding.html
> 
> http://log.amitshah.net/2009/04/re-comparing-file-systems.html

There was some discussion of it on the rsync list, which is where I
learned it is important for NTFS performance on Windows.

By the way, why is fallocate() support being added to libvirt to
improve disk image creation, instead of to qemu-img?

-- Jamie




reply via email to

[Prev in Thread] Current Thread [Next in Thread]