Connect to TAPIS
HARP uses your existing TAPIS access token to list the apps and systems you can use.
App
Sweep
Execution systems and logical queues
from TAPISChoose a TAPIS logical queue, then open it to set what each job should request: cores per node, memory per node, and a time limit. Queue summaries show the maximum values TAPIS permits, not what HARP will request. Add more configurations to profile the same queue with different hardware; each configuration gets its own TAPIS jobs.
Plan
One row per sweep and queue. Each row becomes its own TAPIS jobs.
Results location
When each job ends, TAPIS copies its output here, and the merged profiling CSV is written here too. Filled in from your last sweep or your account; change it if you like.
Copy the results to another location (optional)
When the sweep finishes, the server copies its whole folder (merged CSV, manifest, every job's output) here with a TAPIS file transfer. No account is needed: transfers are not jobs and use your TAPIS login on both systems. You only need permission to write to the folder. The original stays where it is.
Sweeps
Failure details
HARP groups TAPIS and scheduler messages into common causes. Expand a raw error below when reporting it to an administrator.
TAPIS jobs
App
apps with profiling dataTime by run type
SD scaled-down, FS full-scale, test_data held out.
Time by hardware
Median time per run, per system, queue and configuration.
Time by parameter
Median time for each value of every parameter that was varied.
Executions
App
apps with profiling dataSweeps to learn from
Tick the sweeps whose runs the models learn from. Training needs SD, FS and test_data runs.
Data checks
Models to build
Trained on
Where
Trained models
No models built for this app yet.
App
Run
one value per parameterExecution systems and logical queues
from TAPISTick where the run would go. Each available partition gets its own estimate.
Estimates
Each entry, run through the model you selected for its app.