The Free Cluster is available for testing. A draft user guide can be found here (the document will be updated as testing progresses).
The Free Cluster has a number of login nodes which can be used to submit batch jobs and for interactive computing as well, and over 200 CPU cores for batch jobs. Sessions on the login nodes are load balanced and automatically scheduled when you connect to the cluster (no qrsh needed). These login nodes can be used for interactive computing, on similar terms as the interactive nodes in met-cluster, and these login nodes have now more memory than the interactive nodes on met-cluster.
Some popular applications including Matlab, IDL, python, and compilers are already installed and available via the module command. Further applications and libraries can be installed as required.
In batch mode user jobs are isolated and allocation boundaries are strictly enforced. It will prevent oversubscribing resources, runaway processes and will minimize negative impact that incorrectly submitted jobs can have on other jobs sharing the nodes. In this environment, job’s running time will be more predictable, and it will be easier to run larger MPI jobs.
Further improvements include fiber-optic networking for better MPI performance and faster and more reliable connection to data storage (Silver tier is recommended for I/O performance).
More login nodes, compute nodes and some scratch storage will be added as cluster usage increases or more hardware becomes available.
Users of met-cluster are encouraged to migrate their applications to Free Cluster to benefit from better hardware and up-to-date software. At some point we will stop maintaining software on met-cluster, and further, met-cluster will have to be decommissioned. Starting migration now will give us time to install all the software on Free Cluster, and to get ready before met-cluster has to be switched off for technical reasons.