|
From: Jon A. H. <jo...@gm...> - 2012-08-22 16:40:07
|
2012/8/22 Natee Tongsiri <nat...@gm...>: > That's good to know! > > Natee > > PS - for a project that has not been completely ported to debian, there must > be a good reason to choose "dracut" as the ram disk creator for Kestrel huh? > maybe, there's no other like it? The initial ram disk plays an important role on a diskless system since it is the responsible of mounting the root filesystem before the system can start by its own. This root filesystem needs to be writable but the NFS share needs to be read only since it will be shared a long of multiple nodes, so an special filesystem called unionfs is used which allows to stack a tmpfs (a ram based filesystem) on top of the read only system. The problem with the classic initramfs generator (mkinitramfs) is that the code is quite large and complex and difficult to customize. I wanted to add NFS4 support but mkinitramfs lacked this functionality. Then I found that the Red Hat guys where trying to create a new cross distribution, modular and event/parallel (udev) based initial ram disk, and I found that it was really interesting, so I helped porting it to Debian: http://git.kernel.org/?p=boot/dracut/dracut.git;a=search;s=Jon+Ander+Hernandez;st=author The console module is not so important since if booting fails and you end under a rescue shell with busybox IMHO it is not so important not having your regular console fonts and your regular keymap. ;-) In the end it will allow creating interesting modules. I have a proof of concept module which allows to netboot using a wifi connection which could be really interesting for creating quickly other kind of clusters such as videowalls. And that is the history of Dracut and Kestrel. |