The software should be able to run a bench on all (available) nodes invidually, in parallel without the need to wait for one result before running the bench on the following node. Eg.:
# ubench run -b hpcc -w @all
Unclebench should be able to get all available nodes in the job scheduler (excluding automatically the down/draining nodes), and submit all needed jobs itself.
Then, get the results with this command:
# ubench result -b hpcc
node001: <results>
node002: <results>
This result command should also be able to tell if jobs are still pending, if the nodes were down or draining during the run command, and so on.
pierretr
commented
7 years ago
The software should be able to run a bench on all (available) nodes invidually, in parallel without the need to wait for one result before running the bench on the following node. Eg.:
Unclebench should be able to get all available nodes in the job scheduler (excluding automatically the down/draining nodes), and submit all needed jobs itself.
Then, get the results with this command:
This result command should also be able to tell if jobs are still pending, if the nodes were down or draining during the
runcommand, and so on.