Feynman Cluster Main Nodes, Storage, Quotas and Backup Policy
(also see a PowerPoint presentation to Computing Committee)
Main Nodes:
Feynman/Head
Processor: 2x4core Intel Xeon 3.00GHz
Memory: 16GB RAM, 2GB Swap
Usage: Interactive login sessions, for checking emails browsing, editing, et al.
NOT to be used for compute-intensive jobs. Only a front-end for 'submitting' PBS batch jobs.
Phy3/Head2
Processor: Dual Intel Xeon 3.2 GHz
Memory: 2GB RAM, 2GB Swap
Usage: Serves as a backup to feynman. Helps reduce downtime in case of a feynman crash.
NOT to be used for compute-intensive jobs NO interactive logins
Compute Nodes node001-node007
Processor: 2x4core Intel Xeon 3.00GHz
Memory: 16GB RAM, 2GB Swap
Usage: Compute-intensive batch jobs (running for more than 1 hour).
Jobs submitted via PBS actually 'run' here. NO interactive logins
Compute Nodes node023
Processor: Dual-core Dual Intel Xeon 3.2 GHz
Memory: 2GB RAM, 2GB Swap
Usage: Compute-intensive batch jobs (running for less than 1 hour).
Jobs submitted via PBS actually 'run' here. NO interactive logins
Compute Nodes node024-node069
Processor: Dual-core Dual Intel Xeon 3.2 GHz
Memory: 2GB RAM, 2GB Swap
Usage: Compute-intensive batch jobs (running for more than 1 hour).
Jobs submitted via PBS actually 'run' here. NO interactive logins
Web Server wwwphy aliased to physics.princeton.edu
Processor: Dual-core Dual Intel Xeon 3.2 GHz
Memory: 2GB RAM, 2GB Swap
Usage: Primary Web Server of the department running LAMP service (Linux Apache MySQL PHP)
Web Server wwwphy2
Processor: Dual Intel Xeon 2.8 GHz
Memory: 2GB RAM, 4GB Swap
Usage: Backup Web Server of the department running LAMP service (Linux Apache MySQL PHP)
User and Group disks, quotas and backups
In order to optimize the disk usage in the cluster, we provide:
- 100 MB of personal quota on /home disk. User mail and personal web space is accounted in this quota. Some quota extension provided on need-based requests submitted by users.
Back up policy: Local disk snapshot every Sunday, incremental on weeknights, and TSM. - 4 GB of group quota on /group/<group> disk, some quota extension provided on need-based requests submitted by group tech contacts.
Backup policy: Local disk snapshot every Sunday, incremental on weeknights. - 1 TB on /scratch/ disk shared by all users, no quotas, file retention guaranteed for a minimum of 1 week, cleaned up every night, oldest files deleted first to keep sufficient (current target 50%) space always available for scratch work.
Backup policy: No back ups - 25 GB on /tmp of every node (total 1.5 TB) for batch jobs, must be cleaned up by the user soon after job terminates. Users are HIGHLY encouraged to use this space for jobs doing lots of disk I/O, provides fastest access, reduces the load on NFS and thus chances of disk server crashes.
Backup policy: No back ups - /home, /group and /scratch are mounted on the compute nodes, so batch jobs can use them.
Beyond this, groups buy their own disk servers. We help the groups with buying, installation, administration and integration into the cluster. Currently, we have the following group-owned storage:
Mount Point Capacity Contact
----------- -------- -------
/mnt/act 10.0 TB Lyman Page
/mnt/act1 2.1 TB Lyman Page
/mnt/act2 2.1 TB Lyman Page
/mnt/astpart 10.0 TB Christiano Galbiati
/mnt/biotheory 3.5 TB Curt Callan
/mnt/boone 1.0 TB Peter Meyers
/mnt/cmb 1.2 TB Lyman Page
/mnt/cmcomp 1.4 TB Ravin Bhatt
/mnt/cmexp 1.4 TB Jason Petta
/mnt/cms 10.0 TB Chris Tully
/mnt/cms1..12 6.5 TB Chris Tully
/mnt/cosmos 0.3 TB Uros Seljak
Group web space and permissions:
Since /group/groupname/www/ is a group-shared directory, all the groups asked that any one should be able to create his/her own files/folders and others should not be able to modify for a short period (one day) but later, others should be able to update the group's website. To achieve this goal we changed the ownership of /group/groupname/www to group_contact:groupname and created a night cron job that:
- Changes the ownership of all the files to the same as the owner of /group/groupname/www/
- Changes mode of all the subdirs to a+x
- Changes mode of all files to g+rw,o+r
