seisflows.system.wisteria
Wisteria is the University of Tokyo Fujitsu brand high performance computer. Wisteria runs on the Fujitsu/PJM job scheduler.
Note
Wisteria has two node groups, Odyssey (compute nodes) and Aquarius (data/learning nodes w/ GPU)
Odyssey has 7680 nodes with 48 cores/node
Aquarius has 45 nodes with 36 cores/node
Aquarius also contains 8x Nvidia A100
Note
Wisteria Caveat 1
On Wisteria you cannot submit batch jobs from compute nodes and you cannot SSH from compute nodes (Manual 5.13), so the master job must be run from the login node or the pre-post node (Manual 5.2.3)
Note
Wisteria Caveat 2
On Wisteria, the login node Conda environment is not inherited by compute nodes, so it requires custom submit and run script which first load the correct modules, and then run the corresponding script
Note
Wisteria Caveat 3
On Wisteria, command line arguments for the submit and run script, normally input like ‘–key value’ interfere with the batch submission cmd pjsub. So instead we use the pjsub ‘-x’ flag which allows us to set environment variables. We use these in place of command line arguments
Classes
System Wisteria |
Module Contents
- class seisflows.system.wisteria.Wisteria(group=None, rscgrp=None, gpu=None, **kwargs)
Bases:
seisflows.system.fujitsu.FujitsuSystem Wisteria
University of Tokyo HPC Wisteria, running Fujitsu job scheduler
Parameters
- type group:
str
- param group:
User’s group for allocating and charging resources. In the pjsub script this is the ‘-g’ option.
- type rscgrp:
str
- param rscgrp:
the resource group (i.e., partition) to submit jobs to. In the pjsub script this is the ‘-L rscgrp’ option. Available `rscgrp`s for Wisteria are:
debug-o: Odyssey debug, 30 min max, [1, 144] nodes available
short-o: Odyssey short, 8 hr. max, [1, 72] nodes available
regular-o: Odyssey regular, 24-48 hr. max, [1, 2304] nodes available
priority-o: Odyssey priority, 48 hr. max, [1, 288] nodes available
debug-a: Aquarius debug, 30 min max, [1, 1] nodes available
short-a: Aquarius short, 2 hr. max, [1, 2] nodes available
regular-a: Aquarius regular, 24-48 hr. max, [1, 8] nodes available
share-debug: Aquarius GPU debug, 30 min max, 1, 2, 4 GPU available
share-short: Aquarius GPU short queue, 2 hr. max, 1, 2, 4 GPU avail.
- type gpu:
int
- param gpu:
if not None, tells SeisFlows to use the GPU version of SPECFEM, the integer value of gpu will set the number of requested GPUs for a simulation on system (i.e., #PJM -L gpu=`gpu`)
Paths
- __doc__
- submit_workflow
- run_functions
- group = None
- rscgrp = None
- gpu = None
- _rscgrps