Orion Environment Variables
Orion sets several environment variables that can be accessed in a cube. The hardware given to each allocation of a cube could be different depending on what was required and the instance type.
BLIS_NUM_THREADS
The number of CPUs available for use. Used by BLIS (and by PyTorch).
CUBE_TYPE
One of the five fundamental cube types or a group variant of that type:
“sink_cube”
“source_cube”
“compute_cube” or “group_cube”
“parallel_cube” or “parallel_group”
“hyper_cube”
LD_LIBRARY_PATH
By default, set to “/runtime/lib64”, which includes libraries necessary for using GPUs.
MKL_NUM_THREADS
The number of CPUs available for use. Used by Intel’s MKL (and by PyTorch).
OPENBLAS_NUM_THREADS
The number of CPUs available for use. Used by OpenBLAS (and by PyTorch).
OPENMM_CPU_THREADS
The number of CPUs available for use. Used by OpenMM.
ORION_ALLOC_INDEX
An integer uniquely identifying an allocation in a parallel cube or group.
ORION_CPU
The number of CPUs available for use.
ORION_CPU_LIST
A comma separated list of the cpu set assigned to the allocation.
ORION_DISK
The amount of temporary storage in megabytes available for use.
ORION_GPU
The number of GPUs available for use.
ORION_INSTANCE
The instance type providing the resources to the allocation.
ORION_JOB_ID
A unique identifier given to each job.
ORION_MEM
The amount of memory in megabytes available for use.
ORION_NUM_PROCS
The total pids-limit value for a cube.
ORION_PROJECT
The ID of the project the job is running in.
ORION_TAGS
A list of comma separated strings indicating unique features of an instance. For example “A10” indicates that the allocation has access to an NVIDIA A10 GPU.
PATH
By default, includes “/runtime/bin”, which includes binaries necessary for using GPUs.
PYTHONUNBUFFERED
Set to
1so that log messages appear more quickly.
TMPDIR
Path to a writable temporary directory. In Orion, this directory is
/scratchand is private to each cube allocation. See cube file systems.