In the single workstation case, a greatly simplified configuration is possible. See Failed to launch! 255 Host key verification failed? for an example.
Briefly summarized (and omitting important details/steps), one could in the single workstation case:
- define the master hostname as localhost
- define the worker hostname as localhost
and consequently bypass a number of issues related to domain name resolution, remote port access and ssh logins.
To make, on the other hand, the CryoSPARC instance expandable for the future addition of separate GPU workers, one has to ensure:
- password-less ssh logins from the master host to GPU nodes
- GPU nodes must be able to correctly resolve the master’s hostname and contact the master host via various network ports