From djplessinger at sbcglobal.net  Wed Jan 12 18:10:15 2005
From: djplessinger at sbcglobal.net (Derek Plessinger)
Date: Tue Nov  9 01:14:28 2010
Subject: [scyld-users] Please help, slave node will not boot.
Message-ID: <20050113021015.77923.qmail@web81608.mail.yahoo.com>

Hi, I hope you can help.  Let me first say that I am
new to clustering and linux.  I sucessfully installed
scyld on the master and it start's with no errors.  I
then boot up a node using the cd, sometimes it
displays sending rarp requests and other times it says
code at the bottom and displays random letters and
numbers on the screen.  On the master the mac address
appears from the node, I move it to the middle column
and click apply, next the network connection on the
beostatus goes up for a couple of seconds like both
computers are communicating then it goes back to 1kbps
and the node just shows the status of down and nothing
else.  I then forced the daemons to restart by typing
/sbin/service beowulf restart.  Everything says ok
except for it says
Setting Up libraries: 
find: /usr/lib/libgmpi.so.1: No such file or directory
Please help.

From jkrauska at cisco.com  Thu Jan 13 23:38:42 2005
From: jkrauska at cisco.com (Joel Krauska)
Date: Tue Nov  9 01:14:28 2010
Subject: [scyld-users] distcc on scyld?
Message-ID: <41E77702.2010202@cisco.com>

For fun I started trying to setup my scyld-beowulf cluster to run distcc.

I've run across a fundamental problem.

The compute nodes don't have access to any of the compilers.
So gcc can't spawn cc1:
gcc: installation problem, cannot exec `cc1': No such file or directory

I'm fairly certain this is by-design in that the Bproc application 
migration apparently loads locally and then migrates to the compute nodes.

I'm not sure what to think about this.


Is it foolish to allow expand the nfs share beyond just /home and allow 
the compute nodes access to the full binary tree?



I wanted to disect this a bit further and thought I'd run a simple perl 
script.

--------------------------------------------------------------------
#!/usr/bin/perl

print "hello world\n";
--------------------------------------------------------------------

You can't get much simplier than that..

It runs fine on the master node, but fails when trying to run on a 
compute node:
$ bpsh 0 ./testperl
/usr/bin/perl: error while loading shared libraries: libperl.so: cannot 
open shared object file: No such file or directory

So essentially it seems to me that any application or tool that needs to 
dynamically load /anything/ will be difficult to get going under scyld.


Are there ways around this sort of problem?

Thanks,

--joel