Dear new user,
You now have an account on cuppa, CIRA's beowulf cluster. Please log on and change your password ASAP if you haven't already done so.
To log on, go to any node (01 through 24), eg.
> ssh YourUserName@cuppa01.cira.curtin.edu.au
Once you're within the cluster you don't need the suffixes, you can just
ssh cuppa10 etc.
Currently not all nodes are available. The following nodes are down at the time of writing: cuppa02, cuppa04, cuppa05, cuppa06, cuppa12, cuppa13, cuppa20.
If you haven't been added to the cuppa-users mailing list, please subscribe yourself here. Any planned downtimes etc. will be publicised here - ie maintenance, or times when we need all users to avoid certain nodes for eVLBI.
CUPPA has a significant amount of disk space distributed around the nodes, but it is shared, and the VLBI correlator has priority. If you need disk space for a short period, please speak to Cormac to make arrangements. Similarly if you need to do a large processing job, this should be coordinated with Cormac to avoid impacting correlator operations.
Things you need to know: Cuppa01 is the head node (aliased as cuppahead). Most astro applications are installed on /nfs/apps for cluster-wide access. If there is software you need that isn't in /nfs/apps, please consult with Cormac, Ali, Hayley or Shane before you add anything. Standard Linux packages can also be added by any of those people on request. If you find you DO need to add packages, please let Shane, Ali and Mick know, as they keep a list of packages we require so that if ever a node has to be rebuilt they can do a standard install of packages so all nodes remain more or less equivalent. There is an IDL license available - contact Hayley for details. We don't currently have a license for Matlab on CUPPA.
You have a /home directory. The /home partition on cuppa is shared space and isn't very big, so don't use this for storing any large data sets, just small things like programs, etc. You shouldn't exceed 200 MB of storage in your home area (except in the very short term). If you need to store a lot of stuff, the /data directory is pretty big - though it is independent on each machine, so you'll need to remember where you leave stuff! Be a good citizen and don't hog space you don't need.
There is also /arch, which is raided space, on cuppa03. This is larger storage space, but shared and priority is given to storing VLBI data for correlation - so again, if you find yourself needing space of ~TB to store data etc long term, I'd recommend getting a pbstore account on iVEC ( apply here).
Please Note: NO AREA on cuppa is “secure” - we do not back anything up except the head node's installation information, so it is VITAL that any important information you have on cuppa is backed up elsewhere, such as your own workstation, or secure storage space at iVEC.
To check how busy the cluster is and pick a node that's not overrun at any given time, keep an eye on Ganglia. Cuppa03 is the head node for the VLBI correlator, which is the primary user of CUPPA. Please avoid using this node, especially the local disk space.
There are 5 XRAIDs (disk chassis used for Australian VLBI), these are attached to cuppas 01 through 05. These nodes tend to do have very busy network connections, and cuppa01 experiences a lot of disk changes which makes it slightly unstable, so for most work you may prefer to use nodes 07 through 24.
Nodes 07 through 20 should all be equivalent (cuppa19 is the LDAP server but that should not affect you). Nodes 21-24 are storage servers, but may be used similarly to the other nodes.
Main points: Don't do big processing jobs on cuppa01 - as head node that could potentially slow down everything. Try in general not to use cuppa01 through 05 heavily, as they are regularly used for data transfers. Limited RAIDed (though not “secure”) space is on cuppa03. Please do not use the /data disk space on cuppa03 as that is reserved for the VLBI correlator.
If you have any questions, mostly we maintain the cluster ourselves, so talk to Cormac, and for IT support - hardware issues etc, talk to Ali, Shane or Mick, and cc to scienceithelp at curtin.edu.au or service.desk at curtin.edu.au. If you get in anyone's way, it's likely to be Hayley or Cormac who unceremoniously kill your job - that doesn't happen often though!
Finally, if you use facebook, you might like to become a fan of the Curtin University Parallel Processor for Astronomy (CUPPA), where we sometimes send out more fun updates about the cluster or CIRA in general :)