[AG-TECH] linux capture implementation w/NFS

Matthew Wolf mwolf at cc.gatech.edu
Wed Dec 11 11:01:52 CST 2002


Hi,

Terry & Bob suggested I write this up when I was chatting with them at
SC, so here goes.  This is part description of my personal hack to AG
1.x, and part suggestion for AG 2.0.  It falls on the borderline of
excessive unix techno-weenie, so please bear with me.  (Hey, it's the
holidays... :)

Because of the way linux is deployed inside our college, the /usr/local
directories are all nfs mounted.  This is very handy in terms of
versioning control (and security updates, etc etc), but it does mean
that the default configuration location of /usr/local/ag/lib won't
work.  So here's a (shortened) list of the hacks/manipulations I had to
make.

1) I created a directory /etc/ag on each of the intended capture
machines.  /etc is pretty much always a local file system, so it was a
reasonable choice.  The node-config, machine-config, and video-config
files go in there.

2) In arm-eventlistener and vrm-eventlistener, you switch the
system_configs directory to /etc/ag.  This way, if you try to run
vrm-eventlistener on a machine that hasn't been configured with /etc/ag,
it doesn't do anything (other than complain about files not being
found).  

3) Also in vrm-eventlistener you need to add an additional argument to
the final command.  Namely,
   $JAVA ag.VideoResourceManager --eventURL $EVENT_SERVER_URL 
   --cacheURL $CACHE_URL --config=$system_configs/video-config $*


You have to go through the pain of actually doing the nfs server
configuration from here, but otherwise that's it.  You need to be sure
that you have nfs file locking turned on, otherwise rat will fail
silently (it tries to aquire a lock on the file .mbus in your home
directory).  You can take this further to do a number of other nice
things -- trapping error messages in the eventlisteners so that you can
tell people who to contact if they want to use their machine for the
access grid, adding configuration switches so you can move machines
between different display nodes, etc.

The really nice thing is I can bring up a capture box pretty much
instantly -- I have frequently had someone come in for a demo and say
"but I want a camera that can point at X" where X is someplace in the
back corner and/or another room.  The fact that it takes so little
overhead to plug in a camera & bring it online has saved the day several
times.

Matt

-- 
                                        
        -- Matthew Wolf 
        -- Research Scientist, IHPC Laboratory
        -- College of Computing, Georgia Tech
        -- mwolf at cc.gatech.edu



More information about the ag-tech mailing list