Servers #15
Description
Ideas on how to handle different servers and general design of interaction with server/computer/cluster.
Ideas/features
- Each machine should have a capabilities struct, detailing scheduler + account ?, nodecount, cpus/node etc
- By once connecting to the server the central hub can ask for the capabilities and store them
- The connection method also determines how transport should happen i.e. ssh -> scp (initially only this supported)
- Initially only slurm or bash scheduling supported
- Would be nice if codes can be abstracted in a sense that the right codes etc are automatically selected when submitting to a given computer
Questions
- Do server-side jobs and commands for cp/mv submit etc get submitted to the server-side daemon, or straight through ssh commands