Originally Craigslist donated a bunch of computers to BSOE in the summer of 2009. We built a computer cluster out of it, and are allowing students, staff and faculty at UCSC to use the cluster . That original hardware is now retired.
The new cluster consists of one head node, and 6 compute nodes with a total of 240 cores to execute processes on. There are 2 queues all.q for long running jobs, and small.q for shorter jobs. Individuals can run 50 jobs at anytime, but queue up as many as they need.
The new clusters name is campusrocks2.soe.ucsc.edu
Campusrocks uses Cruzid Blue to login. You must have set your password since 10-15-11 for this to work. Visit the Cruzid web page to set your password.
Once you have a password, you can log in to the system via SSH at campusrocks2.soe.ucsc.edu.
It has been given a new 5 year life cycle, and will be unplugged at the end, unless
more funding appears to rejuvenate it. Its new drop-dead date is January 1st, 2018.
If you wish to purchase more nodes for the cluster, please put in a ticket help@ucsc.edu
It came about based on an idea ISSDM had to build a campus cluster,
and allow others to use it. The idea is have a shared resource, but
allow those doing Computer research to be able to look at REAL file
systems, to see how people use the file systems. So file system data
will be available for both ISSDM and SSRC, and other units that need
access to it.
Regarding the queuing software, we are using SunGrid SGE and its documentation is available on the cluster also.
http://campusrocks2.soe.ucsc.edu/roll-documentation/sge/6.0/
qsub --> Submits a job (create a shell script, then run qsub shellscript)
qdel --> Delete a job
qlogin --> Interactive login
qstat --> See the status of jobs in the queue
qmon --> Gui
NOTE: The web site is locked to on campus access only.
Compile the code with mpicc
/opt/openmpi/bin/mpicc
sample c code is in /opt/mpi-tests/src
And then a shell script like this (mpitest16):
#!/bin/csh
unsetenv SGE_ROOT
/opt/openmpi/bin/mpirun -np 16 -machinefile $TMPDIR/machines /opt/mpi-tests/bin/mpi-ring
Finally submitted to the sungrid queue like this:
% qsub -pe mpi 16 mpitest16
qhost shows the load averages of each of the exec hosts
qstat -g c gives a count of number of jobs running on each queue
The Small queue is for jobs that will not run for a long time, there is a 72 hour wall clock limit and and 800 hour CPU limit (if you do multi-threaded operations)
you can see queue configurations with the command qconf -sq small.q
The small.q currently has 2 boxes dedicated with 48 processors each
(The all.q has 4 computers with a total of 144 processors)
We will load RPMs that are in the yum repository for the OS we are running, or you can compile code yourself in your home directory.
Put in an ITRequest ticket for known RPMs.
The /campus directory (your home directory) has an rsync done daily. Look in the /backups directory to find them. We run snapshots in the /backups directory cd /backups/.zfs/snapshots (we keep 4 daily snapshots, and 3 monthly snapshots)
The original donation by craigslist is almost all gone. The only pieces still in place are:
1.) Head node
2.) The boxes the fileservers
For the rest:
SOE purchased new network switches
SOE purchased the 2 48 core boxes
PBSCI purchased the 2 64 core boxes
the 2 8 core boxes were old campus VM servers.
If wish to contribute new hardware, please contact us.