If you plan to install the QCG-Computing service on a separate machine (for performance or security reasons) you must assure that:
- the machine is a submit host of the batch system, what further implies that:
- the machine share account information with the whole cluster (e.g. using LDAP or NIS),
- the batch system/scheduler client packages are installed,
- the QCG-Computing host is an authorized submit host
- the host mounts some shared file system that is available on the worker node machines (however it does not have to be a HOME filesystem)
Further sections gives some configuration hints for various batch system
Torque and Maui/Moab
- add qcg machine as submit host
> qmgr Max open servers: 4 Qmgr: set server submit_hosts += qcg.your.domain set server submit_hosts += qcg
Note: You may need to add the QCG-Computing host in the /etc/hosts.equiv file.
- the machine mounts /home filesystem of the cluster. Alternatively (or in addition) you can mount any other file system that can be used by any user (e.g. the scratch filesystem).
- the PBS_SPOOL/server_logs directory is exported in read-only mode (e.g. using NFS or Lustre) from Torque server to this machine (this share must be mounted locally as PBS_SPOOL/server_logs). The Torque DRMAA library for scalability reason can be configured to use batch system logs to get notified about jobs state changes.
- the PBS_SPOOL/server_priv/accounting directory is exported in read-only mode (e.g. using NFS or Lustre) from Torque server to this machine (this share must be mounted locally as PBS_SPOOL/server_priv/accounting). This file is used to determine exec hosts of short running (needed by BAT accounting).
- the machine is administrative host of the Maui Scheduler, what further denotes that:
- The maui-client package is installed.
- On the QCG-Computing machine modify the SERVERHOST section of the /var/spool/maui/maui.cfg file. It must points to the machine where the maui daemon is running.
- On the Maui server machine modify the ADMINHOSTS section of the /var/spool/maui/maui.cfg file and add the QCG-Computing host to the list of admin hosts.
Final Notes
Note 1: Although it is not mandatory we recommend you to restrict SSH access to the service machine by adding the following line to the /etc/sshd_config (remember to reload sshd deamon):
AllowUsers qcgadmin1 qcgadmin2 siteadmin otheradmin
Note 2: QCG service node should be treated as any other server machine in your cluster, i.e. don't forget about regular packages updates, general node monitoring used at site, security audits.
Attachments
- DeploymentConcept.png (65.4 KB) - added by mmamonski 13 years ago.