So far I've found very little information on the scheduling policy, resource management policy of Azure HPC Scheduler.
I would appreciate any kind of information regarding some of these questions:
- What scheduling policy does a Head Node use to scatter jobs to Compute Nodes?
- Does Azure Scheduler use prior information about the jobs (compute time, memory demands ...) ? If 'yes', how it gets this information?
- Does Azure Scheduler split a job into several parallel jobs on one Compute node?
- Does it have any protection from Compute Node failures? (what it does when a compute node stops responding)
- Does it support addition/subtraction of Compute nodes?
- Is it possible to cancel a job? P.S. I'm aware of the MSDN resource Windows Azure HPC Scheduler. I found only information of how to use this Scheduler but almost nothing about how it works inside.