How do I get an account on a cluster?
If you already have an DSC account, access can be requested through SCUM. These requests will be sent to the cluster owner for approval. If you do not already have an DSC account, you must first request a DSC sponsored account. To request a DSC sponsored account, go to SCUM, click on "Request a DSC Account", select "Sponsored Account", then select the DSC faculty member that will act as your account sponsor. Once the account sponsor approves your request, you can request cluster access.
How do I log into the cluster?
Cluster resources (compute nodes, etc.) can only be accessed through the cluster headnode. Inside the DSC network, cluster headnodes can be accessed using ssh. For example, to log into a cluster called phoenix, type:
If you are outside the DSC network, you must first log into PAMD to get past our firewall:
How do I compile programs?
Use the cluster headnode to compile programs intended to run on that cluster. On the headnode, your default path will be set to allow access to all compilers available on that system. For certain proprietary compilers, licensing issues will limit the number of simultaneous compilations allowed. If a compilation job fails because there are no available licenses, be patient and wait for other users' compilation processes to complete, then attempt your compilation again.
How do I run programs?
The headnode must be available to provide essential services to other cluster nodes, so it is imperitive that resource-intensive programs are never run directly on the headnode. Instead, all programs should be submitted via the cluster's queuing system. Although ssh may be allowed to compute nodes, please refrain from using this ability to run jobs, because this prevents the scheduler from optimally managing the cluster's resources.
How do I log into a specific node?
In the event that it is necessary to direcly access a specific cluster node interactively, please use the cluster's queuing system.
qlogin -q [queue-name]
you will be given a login on a node of that type. If you need to log into a specific node you can use the form
qlogin -q [queue-name]@[machine-name]
What are the differences between the MPI implementations?
In general, most MPI implementations provide the same functionality, but in slightly different ways. OpenMPI provides many utilities that make it more portable and dynamic than the others, making it the suggested platform for new software development. Occasionally programs require features that are specific to another MPI implementation. Other MPI implementations are available on certain clusters, but often are often less scheduler-friendly, and have been known to leave processes running if a job dies unexpectedly.