Setting Up a Torque Cluster in StratusLab

This tutorial will guide you through setting up a Torque Cluster using a pre-cooked image from the StratusLab marketplace. You can either use the stratuslab command line client or a pre-configured public SlipStream project in order to deploy the cluster on StratusLab.

The marketplace image to use is: AViUGHkhd5NytPAKnIp03hnsVWG. First, we use the stratus-run-cluster command in order to create a cluster with passwordless ssh between the nodes (using the ssh-hostbased option):

$ stratus-run-cluster --ssh-hostbased -n 3 DiIc0xtLroMk5PxBQeldQmclkvq

This will set up a cluster with a master and 2 nodes. Once the cluster is launched and configured, login on the master node as root and run the script that sets up the Torque Cluster providing the hostnames of the worker nodes, e.g.

$ ./install-torque-with-maui.sh vm002.grnet.stratuslab.eu,vm004.grnet.stratuslab.eu

First you need to create a SlipStream account. Once you login to SlipStream, select the TorqueCluster public project that is already configured to use the torque enabled image. The "cluster" deployment it includes (found under project “Deployments”), creates a cluster with a master and 2 worker nodes. You can change the number of worker nodes by adjusting the multiplicity of the worker nodes accordingly: in the cluster deployment select the “Nodes” tab, press edit, modify the multiplicity field of the “worker_node” node and save the changes. Finally, you can create the cluster by pressing run. In this case, the configuration script is automatically run and the cluster ready to use, as soon as the deployment is finished. Note that in this configuration the root user is not allowed to submit jobs, you can use the linux user “torqueUser” (for whom passwordless ssh is setup) in order to use the torque cluster.

Here are some simple commands to test the configuration:

qstat -q # shows the job queue
pbsnodes -a # shows the worker nodes
qsub -V -l nodes=worker.node.hostname -I # gets an interactive shell to a worker node
echo sleep 10 | qsub # submit some dummy jobs
qstat # list the queued jobs