[EMAIL PROTECTED] wrote:
>
> Ken,
>
> Sounds pretty darned cool.
>
> What is it that you are rendering with your farm ?
Its a farm for Fox News in NYC. They are rendering frames of 3D
animation with Alias|Wavefront Maya. We didn't want to have to maintain
5 individual machines (RedHat updates, Maya updates, etc), so I decided
to take a crack at LTSP using one as a server/rendernode and the other 4
as clients using PXE (no Etherboot) and local app support (Maya, NTP,
etc). I had actually implemented this with a local disk for swap, but
we decided to put the money into more memory instead, since swapping of
_any_ kind just reduces the performance.
As long as I have your ear, do you have any ideas on how to get dhclient
to exit after it receives the info that it needs, so that the ramdisk
(/oldroot) can be unmounted?
One other thing: I have noticed that my LTSP client machines seem to
report less physical memory via /proc/meminfo than the server, even
though they all have the exact same hardware config. Do you have any
idea why this is? Is this something with the kernel, a ramdisk, tmpfs?
If I boot one of the client machines with the server disk, then I see
more memory reported, so its definitely something to do with the LTSP
client machines.
>
> On Fri, 27 Dec 2002, Ken Murchison wrote:
>
> > I didn't see this mentioned in the archives, so I thought I'd share it.
> >
> > In the process of building a LTSP-based renderfarm for a customer, I had
> > to compile my own kernel from the RH 7.3 sources (for reasons that I
> > won't get into). As you may or may not now, RedHat has a bunch of their
> > own patches to the 2.4.18 kernel, so the Swap-over-NFS NFS patch does
> > not apply cleanly. In order to avoid porting the patch to the work with
> > the RedHat kernel, and to avoid using any third-party patches at all, I
> > started looking for another way to Swap-over-NFS. Here is what I've
> > done:
> >
> > - Compile the kernel for the workstations to include support for loop
> > devices (CONFIG_BLK_DEV_LOOP=[y|m])
> > - Copy your server's /sbin/losetup to /opt/ltsp/i386/sbin
> > - In /opt/ltsp/i386/etc/rc.local, change:
> > swapon ${SWAPFILE}
> > to:
> > /sbin/losetup /dev/loop0 ${SWAPFILE}
> > swapon /dev/loop0
> >
> > The performance of swapping via the loop device might be slower than
> > using the kernel patch (I haven't done any comparisons), but since I'm
> > only doing this as a last-resort fallback for my render nodes (each has
> > 2GB RAM), I'm not really worried about it).
> >
> > After making this change, and using tmpfs instead of a ramdisk for /tmp
> > (my render nodes need more than a few MB of temp space), I couldn't be
> > happier with LTSP.
> >
> > Thanks for the great work!
> >
> > Now if I could only figure out how to get dhclient to exit gracefully so
> > /oldmount gets unmounted...
> >
> >
>
> --
--
Kenneth Murchison Oceana Matrix Ltd.
Software Engineer 21 Princeton Place
716-662-8973 x26 Orchard Park, NY 14127
--PGP Public Key-- http://www.oceana.com/~ken/ksm.pgp
-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_____________________________________________________________________
Ltsp-discuss mailing list. To un-subscribe, or change prefs, goto:
https://lists.sourceforge.net/lists/listinfo/ltsp-discuss
For additional LTSP help, try #ltsp channel on irc.openprojects.net
