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

Re: Using live-boot with mainline kernel



Hi

On Friday 18 March 2011, Daniel Baumann wrote:
> On 03/18/2011 01:57 PM, David Kuehling wrote:
> > As mentioned before I'm currently trying to make live-boot work for a
> > powerpcspe embedded system.  Currently I'm testing this with a fat
> > partition on a usb hard-disk attached to the system.
> 
> unionfs is going to disappear really soon in 3.x; unionmount support is
> not yet in kernel mainline, there's experimental patches against
> live-boot available[0] (and it's a quite high priority to get them
> merged into live-boot soon).
> 
> or in other words: you will have to patch the kernel, either for using
> aufs, or by using unionmount.

Fedora uses a devmapper based approach for its live media, which 
doesn't require any kernel modifications. The basic idea for this is 
to extend a filesystem loop image on tmpfs backed extents (writable 
loop image as a sparse file) in RAM.

Performance of this devmapper approach appears to be slightly better 
than aufs, but it also has some drawbacks, namely the filesystem 
overhead for the required loop image (~about 35-50 MB for an ISO 
image), having to define a fixed maximum size for the sparse file at
boot time and larger RAM requirements, but it does work with pretty 
much any unpatched mainline kernel with devmapper support enabled.

We added this functionality to aptosid some time ago, when aufs' future
appeared to be quite uncertain, but never actually used it in a release
because of the afforementioned disadvantages; it should be relatively
easy to port similar functionality to live-boot though[1].

Of course, VFS based union mounts are the way to go and the ideal/ 
final solution to this problem, unfortunately it appears that its 
development has seriously lost momentum recently and it becomes 
questionable when or /if/ this functionality will be merged mainline.

Another potential alternative would be OpenSuSE's (kiwi) fuse based 
approach, however we didn't actually test that (as we'd like to avoid
'large' userland/ fuse dependencies in the initramfs).

Regards
	Stefan Lippers-Hollmann, aptosid developer

[1]	http://svn.berlios.de/wsvn/fullstory?op=comp&compare%5B%5D=%2Ffll-live-initramfs%2Ftrunk%2Fscripts%2Ffll%4028103&compare%5B%5D=%2Ffll-live-initramfs%2Ftrunk%2Fscripts%2Ffll%4027340
	(plus a small bugfix in r28658)
	Vcs-Svn: svn://svn.berlios.de/fullstory/fll-live-initramfs/trunk
	Vcs-Browser: http://svn.berlios.de/wsvn/fullstory/fll-live-initramfs/trunk/

Attachment: signature.asc
Description: This is a digitally signed message part.


Reply to: