On Thursday 24 November 2005 14.20, you wrote:
> Hi Jimmy,
>       As I stated in some other reply, any bit of information is
> definately deemed useful.
>
> > No duplicate system trees and binaries, just tweak with
> > selective /etc/init.d scripts and such.
>
> Such as...

You just run different services on different machines. Eg not so 
necessary to have nfsd running on a diskless node, or xdm/cups on a 
headless node?

>
> > A few init.d scripts has to be slightly changed from the gentoo
> > originals, along with /sbin/rc and /sbin/functions.sh.
> > The nodes can have different config files. But that is a very
> > minor overhead.
>
> Again, which files and what are the modifications?

/sbin/rc needs to be intercepted to allow for diskless mounting of 
root system over nfs, before the rest of the system tries to wake up. 
Then overload the config areas that need to be different on different 
machines, like runlevels, fstab, etc.
/sbin/functions.sh just adds a function that checks whether it's a 
diskless node or not.
E.g. Intercept checkroot and checkfs on diskless nodes, should not 
happend. Don't try to raise eth0, which has already been done by 
dhcp/kernel,  and make sure you don't take it down. lol.
Don't unmount / nfs on diskless nodes when shutting down, right?
This kind of things, I'll send you the files.

>
> > I have posted on this topic before, but there hasn't been much
> > interest. If you want to try it then contact me and I'll see if I
> > can whip up a quick terse description. Then we can flesh it out
> > as we go. But I don't have time to actually write something nice
> > for quite a while, unfortunately.
> >
> > Harebrafolk
> > Jimmy
>
> Ah, see, that is where I come in. I _have_ to build this cluster
> and it's part of my mandate to document this, and the document is
> supposed to be published in a way or another. I definately intend
> to have this built as a "Classic" Beowulf with PXE booted nodes,
> each with local disks for local caching of large amounts of data to
> be processed.
>
> So, if you don't have time to describe the changes and all, just
> send me some or all of the files which required modifications, I'll
> attempt to figure it out with diff and document this.
>
> Thanks,
>
> Eric Thibodeau

Ok, sounds like you have your work cut out for a little while.
I'll contact you off list about how I can upload the stuff to you.
Also, take a look at:
http://www.gentoo.org/doc/en/diskless-howto.xml
http://forums.gentoo.org/viewtopic.php?t=54293
http://www.gentoo.org/doc/en/altinstall.xml
http://www.gentoo.org/proj/en/cluster/hpc-howto.xml
http://www.gentoo.org/doc/en/openmosix-howto.xml
http://www.gentoo.org/proj/en/cluster/
http://oss.linbit.com/csync2/
http://www.gentoo.org/proj/en/cluster/demo-lwe.xml
(I'm not sure what you want it for, so I can't tell if any of the 
above is of interest)

But: To me a "classic" beowulf is just a bunch of dedicated homogenous 
machines with mirrored system images and mpi/pvm and batch queuing. 
Why the hassle of setting up a more complex cluster system with pxe 
booting? The classic stuff can be solved with more or less just an 
rsync script? If you already have the dedicated homogenous hardware, 
why not save a lot of time, and do what everyone else has been doing 
(tried and true, eh?) for many years, and for which there are a tons 
of available systems and good documentation?
I'm just curious. After all, I ended up rolling my own solution since 
I couldn't find anything that suited my needs at the time.

Harebrafolk
Jimmy
-- 
[email protected] mailing list

Reply via email to