cleaned up build.sh, fixed gennet run.sh, updated gennet

[0xFugue]

• cleaned up build.sh; kurtosis installation is now portable
• cleared run.sh for gennet corner cases
• added node type distribution to CLI
• added port_shift and log_file; tomls continue to be per-node

closes #80 , #69 , partly solves #84 ,

[0xFugue]

I think we can go with this for now, and I am not sure but I think I do not need port-shift information in gennet, as I will handle that in kurtosis with some kind of batch-adding-service.

I mean, there is no point on doing 1000 port shifts if they are in batches of 10 by container for example. Tomorrow I will give you better feedback about this,

I would suggest not to do the packing in Starlark. The container_size and node type will non-trivially interact with subnetwork: With Starlark limitation, containers must be packed with just one subnetwork while minimising both subnet fragmentation and container spill-overs. This means subnetworks and container both need to be developed together. Starlark simply needs to read the gennet generated container packing and populate the script for ENTRYPOINT.

The port_shift will max out at container_size, once that is implemented. I chose not to put in a naive container packing because we need more data/clarity on whether if we allow mix of nodes in a container, what are the trade offs etc.

[AlbertoSoutullo]

I think we can go with this for now, and I am not sure but I think I do not need port-shift information in gennet, as I will handle that in kurtosis with some kind of batch-adding-service. I mean, there is no point on doing 1000 port shifts if they are in batches of 10 by container for example. Tomorrow I will give you better feedback about this,

I would suggest not to do the packing in Starlark. The container_size and node type will non-trivially interact with subnetwork: With Starlark limitation, containers must be packed with just one subnetwork while minimising both subnet fragmentation and container spill-overs. This means subnetworks and container both need to be developed together. Starlark simply needs to read the gennet generated container packing and populate the script for ENTRYPOINT.

The port_shift will max out at container_size, once that is implemented. I chose not to put in a naive container packing because we need more data/clarity on whether if we allow mix of nodes in a container, what are the trade offs etc.

No no, maybe I did not explain myself well enough. Gennet will still assign subnetworks to nodes. I only ment the counting of port-shifting.