HPRC simple Cluster deployment

Deployment of simple simulated High Performance Robotic Computing (HPRC) cluster step by step guide. For more information about HPRC please visit here. This guide is to be followed in computers using Ubuntu.

General configuration

---

Software installation (as root or with sudo)

apt-get install openssh-server nfs-common 
apt-get install python-mpi4py openmpi-common openmpi-bin openmpi-doc libopenmpi-dev 

Add hosts to /etc/hosts , one line per host (as root or with sudo)

<IP>  <FQDN>  <Hostname>

Create home directory for HPRC cluster's users (as root or with sudo)

mkdir -p /export/home

Create user (as root or with sudo)

useradd -u <user_ID> -m -d /export/home/<username> -s /bin/bash <username>
passwd <username>
su <username>

Only in the master node: Generate SSH keys for and configure passwordless authentication

ssh-keygen -t rsa
cd ~/.ssh
cat id_rsa.pub >> authorized_keys
chmod 600 authorized_keys

File system

---

File system installation: In a HPRC cluster, the file system can be any technology traditionally used in Supercomputing. For a simple cluster, Network File System (NFS) will suffy. For this example, the folder /export/home will be shared from the master node. Such folder corresponds to the user space for this configuration

Master node

apt-get install nfs-kernel-server  

Edit the file /etc/exports and add the following lines:

/export/home <Network/Mask>(rw,no_root_squash) 

Execute the following commands

service rpcbind restart
/etc/init.d/nfs-kernel-server restart
exportfs

Slave nodes

Edit the file /etc/fstab and add the following lines:

<SERVER_IP>:/export/home /export/home    nfs    defaults,proto=tcp,port=2049    0 0 

Mount NFS shared folder

mount <SERVER_IP>:/export/home /export/home 

ArduPilot SITL and APM Planner 2

---

For ArduPilot SITL please follow the instructions in here

For APM Planner 2, please follow the instructions in here

Traditional supercomputing (HPC) examples

---

Hello world with mpi4py

PI computation with mpi4py

To execute these examples use the following command:

mpirun -np <X> --hostfile <hostfile> python <pythonSoftware>

where,

X = Quantity of MPI parallel processes
<pythonSoftware> The corresponding software
<hostfile> is a text file with the hostnames of the computers in your cluster, e.g. 
  <Hostname1>
  <Hostname2>
  <Hostname3>

HPRC example

---

For an example see swarming motion based on MPI and DroneKit in here