set the channel description: Chat about https://github.com/GoogleCloudPlatform/rad-lab/tree/main/modules/silicon

@Matt Venn @tnt the project should be created (sorry it took so long, things got a little bit wild here as you might have heard)

let me know if you can access: https://console.cloud.google.com/home/dashboard?project=catx-ext-zerotoasic

@Philipp Gühring same for here: https://console.cloud.google.com/home/dashboard?project=catx-ext-libresilicon

I'll update the radlab deployment and share instruction on how to access notebook next week.

image.png

and no way to change my logged in user

@Matt Venn are you logged in with you rad-lab-silicon.dev address?

I would advice to create a new chrome profile for it, it makes it easier not to confuse the console about which account to use.

I have cloned https://github.com/GoogleCloudPlatform/rad-lab/ to https://github.com/thesourcerer8/rad-lab , there I have copied the modules/silicon to modules/libresilicon , renamed the Docker image from silicon-design to libresilicon-design, and added in the required packages, python modules and downloads into the Dockerfile.

Now how do I register the new Dockerfile as a Docker image? Is that automatically done when launching the module with the radlab-launcher?

I tried the radlab-launcher but ran into a permission issue:

And when I try to go https://cloud.google.com/iam I am offered a free trial, but I was expecting the be able to view my permissions

Do I need to change anything else in my module "libresilicon" to make sure that it doesn't collide with reusing ressources of other projects?

I also tried it locally now:

I have now succeeded to get the standard cell generation to the point that it makes sense to run them massively in parallel. I would like to try generating 800 cells in parallel on the rad-lab-silicon.dev environment. I guess that we could try to run maximum 10 jobs in parallel per machine (since some jobs take a few GB of RAM, and some parts of the flow are also parallelized already (but some parts are sequential). Now my question is which scheduling system should I try to use on GCP to get that done? On a single machine I was used to make -j but that wouldn't distribute it on several machines. I think for the Google cloud I should package up everything necessary into a container, and then perhaps use

gcloud batch jobs submit

to submit the 800 individual jobs? I only have to collect all the generated files in the end, so I guess I have to add a bucket or NFS storage where the results are stored. (Should I use the notebooks bucket for that?)

A "gcloud batch jobs waitforallbatchjobstocomplete" command would be nice to wait for all the results in the end.

Ah, now I found how to grant the permissions to myself

Ok, a few permissions later I get this error, it doesn't seem like I am missing a role:

I've got an example of submitting multiple jobs in parallel from another notebook here

It leverage vizier for parameters optimization

I'd be happy to walk you thru the example

Could you make a small video perhaps for the walk thru? I think many more people could make use of that. If you want, we can do it together.