Deprecation of SGE and Torque in ParallelCluster

Versions 2.0.0 - 2.7.0 of ParallelCluster have included SGE and Torque as job schedulers available through cluster configuration options (in addition to Slurm and AWS Batch). SGE and Torque are open source schedulers created and maintained by a community, but for which there has not been active development in some time (2016-SGE, 2018-Torque). To minimize the risk of unfixable bugs or security issues affecting your business processes, we will be sunsetting our support for these two job schedulers in future releases of ParallelCluster.

Customers are welcome to continue using past releases of ParallelCluster if they wish to continue using SGE and Torque. For customers that wish to continue using later releases of ParallelCluster, however, they may wish in some cases to transition their workloads to using either Slurm or AWS Batch as their workload manager. Below we have provided some resources to assist in this transition. In many cases, migrating from one job scheduler to another is a straightforward process that can be accomplished with minimal changes. We have compiled a few resources that could be useful as part of this process:

• SchedMD, the maintainers of the open source job scheduler Slurm, have published what they describe as a Rosetta Stone for workload managers that can be used to translate commands specific to SGE or Torque (in addition to PBS, LSF, and LoadLeveler) into Slurm commands. https://slurm.schedmd.com/rosetta.pdf
• There are also wrappers available for several SGE and Torque commands. More information is available on SchedMD’s website at https://slurm.schedmd.com/faq.html#torque
• Stanford Research Computing Center has prepared a guide to assist their users in migrating from SGE to Slurm. https://srcc.stanford.edu/sge-slurm-conversion
• The University of Georgia’s computing resource center has published its own guide to assist their users in migrating from Torque to Slurm. https://wiki.gacrc.uga.edu/wiki/Migrating_from_Torque_to_Slurm