Skip site navigation (1)Skip section navigation (2)
Date:      Fri, 13 Dec 2002 04:14:24 +1030
From:      Greg Lewis <glewis@eyesbeyond.com>
To:        Andy Sporner <sporner@nentec.de>
Cc:        "Ronald G. Minnich" <rminnich@lanl.gov>, freebsd-cluster@FreeBSD.ORG
Subject:   Re: sharing files within a cluster
Message-ID:  <20021213041424.A18994@misty.eyesbeyond.com>
In-Reply-To: <3DF8B417.9040509@nentec.de>; from sporner@nentec.de on Thu, Dec 12, 2002 at 05:06:47PM %2B0100
References:  <Pine.LNX.4.44.0212120858520.17485-100000@carotid.ccs.lanl.gov> <3DF8B417.9040509@nentec.de>

next in thread | previous in thread | raw e-mail | index | archive | help
On Thu, Dec 12, 2002 at 05:06:47PM +0100, Andy Sporner wrote:
> Ronald G. Minnich wrote:
> >On Thu, 12 Dec 2002, Andy Sporner wrote:
> >>I think this is called PXE.  Pre-XEcution boot environment from Intel.
> >>When you select Network Boot, it hits a DHCP server for an IP address
> >>and then a PXE server for a floppy image.  Then it kicks off the image
> >>just like a FD.  
> >>    
> >
> >The systems I've used have a 32KB limit. So you use PXE to boot the thing 
> >that boots the thing. It's stupid.
> >
> >Also can't boot over things it doesn't know, e.g. myrinet. We boot over 
> >myrinet.
> 
> Most "real" servers this works perfectly.  I know because
> I was involved in building them!  
> 
> My suggestion to you is that if you can't say something
> constructive is that you should say nothing.  That's the
> problem with academic people they just don't know when
> to just keep quiet.  

Ron is making a real world point though, so his posting is both
constructive and relevant.  His postings may be brief, but they are 
to the point and draw on a lot of experience with HPC clusters.

In this case he is spot on.  PXE is useful, but limited.  PXE clients
that I've seen put arbitrary limits on both the size of the image and 
the download time.  I've seen at least one PXE client that will time
out _in the middle of downloading the image_.  Thats right, its getting the
image but it will still time out.  So, we use PXE to download etherboot
and boot into that and then use that to do the actual booting since it 
doesn't have these limitations.  As Ron points out, this is stupid.  PXE
also has no facilities for multicast that I'm aware of so you have to
tftp your image.  The etherboot client we have has a multicast client
built in, another advantage on big clusters where tftp doesn't scale.
As a result we have a preference for NICs with etherboot flashed onto the 
ROM, not PXE NICs.

So, yes PXE is useful.  But get a big enough cluster and it falls in a
screaming heap.

-- 
Greg Lewis                          Email   : glewis@eyesbeyond.com
Eyes Beyond                         Web     : http://www.eyesbeyond.com
Information Technology              FreeBSD : glewis@FreeBSD.org


To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-cluster" in the body of the message




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20021213041424.A18994>