running MPI, and login in locally into slave nodes

Sean Dilda agrajag at scyld.com
Fri Aug 3 15:08:12 PDT 2001


> But what if I need to lets say change a specific file on a particular
> slave node? Also I ran a command bpsh 0 ls -la, I have one slave node up,
> and I got back the exact same list of files as in my home directory on the
> master node. Is this supposed to be like this? Are the slave node exact
> copies of the master node? Also my ping is working but I can't telnet or
> ssh to the slave node.

That is expected because home directories are nfs mounted on the slave
nodes.  If you do that in any directory that isn't in /home, the results
will not match.

There shouldn't be any files on the slave node that you need to modify.
If there are, you can bpcp a new version over.  There is actually less
than 40M  (and less than 10M on the newest release) of files on the
slave nodes, and almost all of what is over there are libraries.  There
are no binaries and only one or two config files.   The point isn't to
ssh or telnet into the slave node, the point it to use bproc to send
your job over there so that you don't have to have a full install on the
slave node.

By having these slaves nodes with almost no files on them, and all the
files being able to be quickly replaced at boot time, it greatly reduces
the amount of work required to maintain the cluster as you no longer
have to make sure a program gets installed on every node and no longer
have to make sure that passwd files and other config files are all
synced up properlly because all these files only have to exist on the
master node.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 232 bytes
Desc: not available
Url : http://www.scyld.com/pipermail/beowulf/attachments/20010803/1a969b44/attachment.bin


More information about the Beowulf mailing list