Overview of Condor Class Ads potentially useful for users.
For an overview of the standard ClassAds in HTCondor see [https://htcondor.readthedocs.io/en/v8_9_3/classad-attributes/index.html]
Requirements = <...
Specify the requirements the batch node needs to have to run your job. E.g.,
Requirements = (OpSysAndVer == "CentOS7")
would require the node to be running with Linux flavour 'CentOS 7'. Multiple requirements can be logically combined in a C-like syntax, i.e., (Requirement_1 > X) && (Requirement_2 == 'Foo'). See also Submitting Jobs
Request_Cpus = <number cores>
For multi-core jobs require the number of CPU cores reserved on the batch node for the job. As currently occupied CPU resources might have to be drained for jobs with large CPU requests, queueing times can be substantially for large CPU requests. See also Submitting Jobs
Request_Memory = <amount of memory>
Request for more memory than the default 2GB to be reserved for your job on the batch node. If the memory request is larger than roughly 2GB per requested core, Condor might have to drain resources to ensure sufficiently free memory, which can lead to longer queueing times and CPU cores being idle. Note that jobs exceeding the memory limit slightly can probably run fine and are not killed assuming that there is always some memory head room - however when exceeding the memory significantly, nodes will kill such jobs to avoid starving of memory.
Request_GPU = 1
Request a NAF node with a GPU card for Machine Learning training etc. .
See also GPU on NAF
+RequestRuntime = <seconds>
Request more (or less) default runtime (in seconds) for a job.
Without this class ad, a job is scheduled to run up to three hours and removed exceeding that run time. This allows for a better utilization of opportunistic resources.
Set the class ad with the approximate runtime for jobs needing longer than the default 3h upper limit. See also Submitting Jobs