Network of SUN cpu's and GRASS

[gcolello at biosphere.Stanford.EDU]

Simon Cox gave good advice on the remote execution of GRASS using  
X-window access between a single "GRASS server" machine and various  
remote machines. We have a similar setup and use that approach all  
the time. We do this both with remote Suns and remote Nexts (using  
the Next coXist X emulator) going to a single Sun "GRASS server" over  
a thinnet. However, as we found out the hard way, each user must  
login to a separate account on the "GRASS server" machine to avoid  
contention for the x0 monitor. To be GRASS functional each account  
must have a path (in $PATH) set to the grass binary directory (for  
example: set path = /home/CGEF/grass/4.0/bin). One drawback to this  
approach is that it causes all GRASS processes to be executed on the  
"GRASS server" machine. This has obvious negative performance  
implications if several users are using GRASS simultaneously.

We also use another method to access GRASS binaries from that  one  
"GRASS server" machine, but GRASS runs on the remote machines CPU's.  
This is done simply by exporting the GRASS binary directory from the  
"GRASS server" to a remote machine. The remote machine of course must  
then mount this exported directory.  This causes the remote machine  
to think that the GRASS binaries are part of its file system and they  
are therefore executed locally. NFS of course serves the binary from  
the "GRASS server" machine. We have found that this method has the  
drawback that exporting can become yet another tedious (and  
confusing) system admin task as new machines are added to the  
network. To minimize this problem we only export GRASS to one  
machine. This machine can then act as a "secondary GRASS server" for  
the X-window approach. This balances the user load and system admin  
work.

Of course this all means that the GRASS binaries are on one machine  
and the network is being used to pass these binaries around as well  
as all the GRASS images from the "GRASS servers" to X-window  
displays.  I suppose network slowdown could eventually result. We  
haven't noticed that yet, but we rarely have more than 3 people using  
GRASS simultaneously, and our setup is fairly new.

Does anyone see any flaws with what I've presented here? We're still  
learning new things every week.

-----------------------------------------------------------------
Greg Colello
Carnegie Institution, Department of Plant Biology
Stanford University
gcolello at biosphere.stanford.edu