Research computing on Microsoft Azure
Run by the Center for Computational Sciences as part of the ECC cluster. You submit from an ECC login node to an Azure queue, the work runs on a machine created for it, and you copy your results back. These pages say what that costs and write the job script for you.
Start here
- Job script builder
Pick a queue, cores and hours; get a complete Slurm script and what the run will cost. - Current Azure prices
The hourly list price of every machine type we use, on demand and spot, East US and Central US, refreshed every hour. - My jobs and credit
Your own jobs and what is left of your allocation (sign in with your UK account). - Approximate current costs and allocations
Sign-in required.
How a job runs
azsync pushon an ECC login node copies your files to the Azure side.sbatch mycase.sbatchsubmits to an Azure queue. A machine is created for the job, which takes three to five minutes the first time.azsync pull <folder>brings the results back to ECC.
The Azure machines cannot see ECC's file system. Credit is counted in core-minutes: 1250 core-minutes is 1 USD on the test queue today.
Feeds
- /data/prices.json — the hourly price feed, as JSON.
- /data/queues.json — the queues, their machine type, limits and rate note, as JSON.
Help
Need help? Open a support ticket. Say which queue and, if a job is involved, the job id;
azjob <jobid> on an ECC login node prints the full report for a job.
No PHI, FERPA-protected, export-controlled or CUI data may be moved to Azure until the UKY security review of this service is complete. There are no backups and no snapshots on the Azure side: the copy that matters stays on ECC.
Center for Computational Sciences, University of Kentucky · Need help? Open a support ticket