wwMpiGanglia

an attempt to automate the installation and configuration of a cluster with warewulf and mpich

Guidelines for Creating Debian 7.11 Linux Cluster

Create USB bootable linux install

First: download debian “wheezy” amd64 netinstall iso from

https://www.debian.org/releases/oldstable/debian-installer/ (Links to an external site.)Links to an external site.

insert usb stick, find out where it is mounted (on Mac):

(for example, my usb shows mounted on /dev/disk3)

ls /Volumes diskutil list

Partially unmount the usb stick:

diskutil unmountDisk /dev/disk3

Write debian image to USB stick (this will take several minutes):

sudo dd if=debian-7.11.0-amd64-netinst.iso of=/dev/disk3 bs=1m

For a simple installation on a single drive, use the following primary partitioning scheme:

/boot (~16 GB, ext4) Swap (2x Physical Memory) / (~200 GB, ext4) /usr (~200 GB, ext4) (Semi-optional:) /home (~500 GB, ext4) /usr/local (depends on how much custom/source installations you’ll have) /scratch (common computing filespace) (Not semi-optional; i.e., mandatory for true production systems:) /tmp (????)

For the installation process, you’ll typically choose eth0 (external NIC) and DHCP to be autoconfigured. Nothing needs to be done for eth1.

Create a root user account and a regular user account. Remember the passwords for both.

Install base system for Debian Linux on master node. Select: Debian Desktop environment, SSH server, Standard system utils

update the base system (as root user) when it reboots, via:

apt-get update apt-get upgrade

download warewulf/mpich installer

then as root

cd /home/user git https://github.com/wesleychildress/wwMpi.git cd wwMpi cp -r * ~ cd ~ ./install.sh

installer will ask several queastions and is a pretty lenghty so grab a cup of coffee and/or beer

default for all is good for now

Your final networking routing can be checked with:

ip route

and should resemble:

default via 136.160.116.1 dev eth0 proto static 10.253.1.0/24 dev eth1 proto kernel scope link src 10.253.1.254 136.160.116.0/22 dev eth0 proto kernel scope link src 136.160.119.40

Verify the appropriate nfs filesystems are being exported by the master node:

showmount -e 10.253.1.254

add first compute node, named ‘n0001’ with the appropriate HW (MAC) address:

wwsh node new n0001 --hwaddr=b8:ac:6f:32:37:08 --ipaddr=10.253.1.1

Note, if you have trouble booting compute nodes over tftpd from here, it may be that the switch has bad arp info - so, restart the switch

Check that nfs mounts are working on compute nodes, and that they are mounted

and preserving file/directory permissions (as root):

ssh n0001 df -k ls -ltra /home

Examine ownership and permissions of mounts, including /home

Create Non-root (non-privileged) user account, then setup for ssh with password-less login:

su (non-root username)

cd echo -ne '\n' | ssh-keygen -t rsa

(enter no passphrase)

cd ~/.ssh cp id_rsa.pub authorized_keys cd yes "" | ssh n0001

(answer yes)

yes "" | ssh $masternode

(answer yes)

exit exit ssh n0001 ssh (masternode) exit exit exit

No prompt/password should happen for the second ssh

NOTE: If you have a kernel upgrade in this, you may need to update the bootstrap image by executing the following commands:

wwbootstrap 3.21.6 wwsh provision set n0001 --vnfs=debian7 --bootstrap=3.21.6-4-amd64

reboot n0001, then ssh into n0001 and check that ntp info is being picked up by the compute

nodes:

ntpq ntpq> peers

manually add new nodes by individually executing each of the following commands:

wwsh node new n0002 --hwaddr=b8:ac:6f:34:b2:fd --ipaddr=10.253.1.2 wwsh node new n0003 --hwaddr=b8:ac:6f:34:62:c7 --ipaddr=10.253.1.3

........... up to .253 master is .254 and dhcp broadcasting on .255

wwsh node new n0016 --hwaddr=b8:ac:6f:32:2e:d4 --ipaddr=10.253.1.16

Make sure all of the new nodes have their hostnames added to the ssh config files by logging # into them manually over ssh, which should prompt for their addition to the files

confirm that pdsh will execute a command in parallel across an initial set of four compute nodes, as a root and as a normal user:

(https://code.google.com/p/pdsh/wiki/UsingPDSH (Links to an external site.)Links to an external site.)

ssh n0001 ssh (masternode) exit exit pdsh -R ssh -w (masternodename),n0001 hostname pdsh -R ssh -w (masternodename),n0001 uname -a

Optional: install rsyslogd on compute nodes if needed:

chroot /srv/chroots/debian7 mount -t proc proc proc/ apt-get install rsyslog wwvnfs --chroot /srv/chroots/debian7 --hybridpath=/vnfs wwsh dhcp update

then restart compute nodes

Make sure /usr/local/lib is in library path on all nodes, perhaps through the addition of

/usr/local/lib to the file /srv/chroots/debian7/etc/ld.so.conf.d/libc.conf file

Reboot all compute nodes:

pdsh -R ssh -w n0001,n0002,n0003,n0004 reboot

Reboot master node

copy hello.c to home directory

cp ~/configFiles/hello.c /home/(username)/hello.c

Compile executable:

mpicc -o hello.out hello.c

Run the executable as non-root user:

mpirun -hosts (masternodename),n0001 -n 12 ./hello.out