4. GPFS cluster setup

The GPFS cluster is managed by escsMaster node and involves escsConsole, escsCore1 and escsCore2 nodes. Physical data storage is realized ona e-sata device attached to escsCore1 node but data are accessible from every node in the cluster and configuration and ACL permissions are managed by the master node.

Package installation is described in the escsMaster document section about Additional software and must be performed on every node composing the cluster. Adding a node to the existing cluster must instead be performed as described in escsConsole GPFS configuration.

The first step is to make every machine root-accessible via ssh from the master node using ssh public key authentication:

[root@escsMaster /]# ssh-keygen -t rsa
[root@escsMaster /]# cp ~/.ssh/id_rsa.pub ~/.ssh/authorized_keys
[root@escsMaster /]# scp /root/.ssh/authorized_keys escsCore1:/root/.ssh/
[root@escsMaster /]# scp /root/.ssh/authorized_keys escsHost:/root/.ssh/

Now we can instruct the master node about how to manage the gpfs cluster:

[root@escsMaster /]# mkdir /root/escsCluster
[root@escsMaster /]# cd /root/escsCluster
[root@escsMaster /root/escsCluster/]# vim cluster.def
escsMaster
escsCore1:quorum
escsCore2
[root@escsMaster /root/escsCluster/]# vim storage.cfg
/dev/sdb/:escsCore1::dataAndMetadata::escsCore1Archive
[root@escsMaster /root/escsCluster/]# mmcrcluster -N cluster-def \
      -p escsMaster -R /usr/bin/scp -r /usr/bin/ssh -C escsCluster.med.ira.inaf.it -A
[root@escsMaster /root/escsCluster/]# mmstartup -a
[root@escsMaster /root/escsCluster/]# mmgetstate -a
[root@escsMaster /root/escsCluster/]# mmcrnsd -F storage.cfg -v no
[root@escsMaster /root/escsCluster/]# mmcrfs /archive escsArchive \
      -F storage.cfg -A yes -B 64K -m1 -n 10 -r 1 -R 2 -v no

Now gpfs should be up and running and the file-systems has been created on the external storage attached to escsCore1.