Overview
In the spring of 2019 we restructured and updated the existing compute cluster infrastructure of the former IFF-institutes. Several individual but similar clusters are joined at storage- and network-level to provide a uniform mode of access and easier access and exchange of data between different machines. A single login-node now serves all compute nodes. Software components have been upgraded from Centos 5/6 to Centos 7, the former Torque queueing system has been replaced the slurm partition manager.
Access to the New Cluster Infrastructure
User access to the cluster's resources is provided through the queueing node iffslurm. Please read the queueing system documentation [1] to find out how to access compute resources.
Do not run any computation- (or heavy compilation) jobs on iffslurm, the node not intended for this! Computation jobs running here will be killed without notice!
You can compile your code on any CentOS 7 machine in the institute or on a reserved compute node.
Additional nodes for accessing resources or development may be added in the future.
| [1] | https://pgi-jcns.fz-juelich.de/portal/pages/slurm.html |
Transition of Compute Nodes to the New Cluster
All compute nodes in the new cluster adhere to a new naming scheme: iffclusterXXYY where XX is a two-digit decimal number identifying a cluster partition and YY is a two-digit decimal number identifying the node within this partition. For example iffcluster0304 is the fourth node in the third partition. Queue names are inherited from the old systems. Below you can find a table mapping old hostnames to new. The transition is work in progress, if your nodes are not mentioned or not yet migrated, we probably were not informed that it is safe to do so.
| Hostname (old) | Hostname (new) | Queue/Partition | migration complete | Infiniband ready [2] |
|---|---|---|---|---|
| iff402c01-14 | iffcluster0101-14 | oscar | yes | yes |
| iff402c15-28 | iffcluster0201-14 | funsilab[1] | no | yes |
| iff003c25-48 | iffcluster0301-24 | th2[ib] | yes | yes |
| iff003c105-108 | iffcluster0325-28 | th2 | no (pending) | no |
| iff597c02-18 | iffcluster0401-17 | topo | no | yes |
| iff597c19-35 | iffcluster0501-17 | funsilab[2] | no | yes |
| iff597c36-51 | iffcluster0601-17 | nanofer | yes | yes |
| iff003c01-24,73-76 | iffcluster0701-28 | th1 | yes | yes |
| iff003c93-100 | iffcluster0801-08 | viti | yes | yes |
| [2] | multi-node MPI-jobs in queues that span several partitions or partitions with mixed nodes need extra care when choosing the queue |
Data Transition from the Old Clusters
Data stored in the /local-directories have been copied to a new server after the compute nodes were logically moved to the new, unified compute cluster. Please see the table below to determine the path to your existing data. Again, if you can't find your data in this table, it has not been migrated yet.
| Cluster (old) | Path (old) | Path (new) [3] | migration complete |
|---|---|---|---|
| iff003 | /local/th1 | /local/th1/iff003 | yes |
| iff003 | /local/th2 | /local/th2/iff003 | yes |
| iff003 | /local/th3 | /local/th3/iff003 | no |
| iff402 | /local/oscar | /local/th2/iff003 | yes |
| iff402 | /local/funsilab | /local/th1/funsilab/iff402 | yes |
| iff597 | /local/funsilab | /local/th1/funsilab/iff597 | no |
| iff597 | /local/topo | /local/th1/topo | no |
| iff597 | /local/nanofer | /local/th1/nanofer | yes |
Data in /SCRATCH-directories is still accessible from the respectives nodes themselves or from /SCRATCH/NODENAME on iffslurm, where NODENAME is a node from the new hostnames listed in the table above.
| [3] | accessible from the compute nodes and iffslurm |