- Download jumbo DVD iso from rocks website
- Setup VM in virtualbox:
- Frontend node: Linux/Redhat x64 with min 1GB RAM, 30GB HDD, 1 optical drive (with rocks dvd iso mounted), 1 Internal Network interface (any name), 1 NAT, boot from optical drive
- Compute node(s): Linux/Redhat x64 min 1GB RAM, 30GB HDD, 1 Internal Network interface (same name as frontend-node), boot from network
- Start frontend-node VM, type
build
when prompted - Select
CD/DVD based rolls
, at leastos
,kernel
,base
. Some recommended rolls areweb server
+ganglia
(monitoring) andhpc
(OpenMPI and MPICH) - Change the default Fully Qualified Host Name (FQHN) to avoid
compute-0-0 already exists
error as described here - Provide the
eth0
,eth1
and gateway address accordingly. If you don't have any idea, try setup below (works for me):# eth0 10.1.1.1 255.255.0.0 # eth1 10.0.3.15 255.255.255.0 # gateway 10.0.3.2 192.168.0.1
- Choose
Auto Partitioning
when prompted (unless you specifically want something else) - Sit back and wait until frontend installation complete
- Login as
root
then runinsert-ethers
on terminal and selectcompute node
- Start compute-node(s) VM and make sure it boots from network. If it's not, try reboot several times. If it's still not working consider redo everything from step 2 (yup, I did retry several times)
- If you have
ganglia
roll selected, you can check and monitor your cluster on http://localhost/ganglia (on frontend)
-
In case you haven't add
hpc
roll. Download hpc iso from website and run this commands on frontend terminal:rocks add roll hpc.iso rocks enable roll hpc cd /export/rocks/install rocks create distro rocks run roll hpc | bash init 6
-
Once reboot done, run this command and reboot compute node(s) to push the change:
ssh-agent $SHELL ssh-add rocks run host compute '/boot/kickstart/cluster-kickstart-pxe'
-
Add new user to run MPI job and sync it to compute node(s)
adduser username passwd username rocks sync users
Mas Yohanes, Itu yang point 9 harusnya insert-ethers ya bukan install-ethers?