n  
 

Introduction to the Nano Cluster

The Nano Cluster currently has 54 compute nodes and 1 frontends with the following configurations:
  • 54 IBM BladeCenter JS20 (n1-n54) compute nodes.
    • 2  64bit IBM PowerPC CPU 2.2 GHz (PPC970FX)
    • 512 KB ECC L2 Cache
    • 2.5 GB (n1-n35) or 4 GB (n36- n54) PC2500 DDR ECC registered SDRAM
    • 2 Broadcom Tigon3 gigabit Ethernet
    • 1 40 GB 2.5" IDE hard disk drive
  • 6 IBM eserver xSeries 346 (884041U) (nanofs1-nanofs6, not available for login)
    • 2 Intel Xeon CPU 3.60GHz
    • 1024 KB L2 Cache
    • 2 GB DDR2 400 RAM
    • 2 Broadcom Tigon3 gigabit Ethernet
    • 6 300 GB 10000 RPM Ultra320 SCSI hard disk drive

Using the Cluster

Access to the Nano cluster is limited to affiliates of the 6 project PI's.  To apply for an account, visit the Account Request Form.

The frontend node, nano.millennium.berkeley.edu, is accessible via SSHv2.

Support & Community

The nano-users mailing list is a forum for community support of the Nano cluster.  All users are added to this list by default but have the option of unsubscribing. 

Community resources:

Torque Batch Queue (PBS)

To submit a script to Torque, run qsub script.sh. Torque defines environment variables and runs your script on the first assigned node; it does not launch processes on multiple cpus.  You may use gexec or mpirun within your script to launch processes on multiple assigned nodes. 

The default queue is batch which contains all nodes. View the output of qstat -a or qstat -q to check the queue status.

MPI

If you are unfamiliar with MPI, please visit the MPI Tutorial. MPI jobs can be run over the gigabit Ethernet.

Compilers

Filesystems

/home/nano/$group/$user is your home directory.
/home/nano/$group/work is group writable scratch space.
/home/nano/$group/software is group writeable for software installation.
/home/nano/$group/software/bin is in your path by default.

/usr/mill contains supported software, including MPI and IBM compilers.
/usr/mill/bin is in your path by default.

/scratch is local world writable storage on each node and should be used instead of /tmp.
/scratch has a 30 day auto-deletion policy; any file not used within 30 days will be deleted automatically.

Nothing is backed up, ever.  Please plan accordingly.

UC Berkeley Clustered Computing - Last modified on 10-Dec-2007 09:18:32 -0800