Redstorm:Harness Setup
From OSR
First boot harness setup
All of these steps are done on rscsmw2. Your install tree should be on an NFS mounted filesystem so that it is accessible to rscsmw2 and can be re-exported to the service nodes. This only has to be done once when you take over the cage from someone else.
- Set `rs_vm` and `cray_bin` as root:
echo attr_mgr \
--set \
--rs_vm cx24y2c0 \
--cray_bin ~/src/full-nightly-20060909/install \
cx24y2c0 \
| sudo su -
- Power cycle the nodes and make sure the L0 comes up:
rs_power --off cx24y2c0
sleep 360
rs_power --on cx24y2c0
sleep 120
ping -c 1 -w 1 cx24y2c0s0
- Init the harness as root. This copies the harness binary to the L0 systems and starts it running.
echo harness --init cx24y2c0 | sudo su -
- Once the harness is up and running, you need to start `wamd` for the cage. I believe it is resonsible for pushing routes to the nodes when they reboot. You should run the `wamd` somewhere that won't die, such as in a screen session. If it dies you will not be able to communication after rebooting nodes and will lose your consoles to the nodes. It produces lots of error messages, all of which are flagged as *FATAL*. You can ignore all of them.
harness --wamd cx24y2c0
- Start the consoles. I wrote a script for our very old system to do this. I'm still using it and it works. There may be a better way.
~tbhudso/src/nightly/hardware/sandia_integrated/scripts/rs_consoles \
-c cx24y2 \
-l /tmp/tbhudso.console.log \
0..31