Large Brain models fail

Description

Related to this forum entry:

Attachments

63

Activity

Pro tip: press M to comment

Anonymous 
December 16, 2023 at 8:40 PM

1

 

1

1

 

 

 

 

 

11

1

 

 

 

 

 

·

Anonymous 
December 16, 2023 at 8:38 PM

1

11

1

 

1

 

1

 

 

 

1

 

1

·

Benedikt Feldotto 
January 2, 2023 at 3:06 PM
(edited)

I could identify three distinct issues:

  1. on Castor VMs large brains cannot be run because loading takes to long, which blocks the CLE and makes it inresponsive. This leads to a timeout from the CLE

  2. the Virtual Coach runs into timeout after 15 minutes on any system. It needs to relogin every time, this is because the OIDC token expires

  3. on NEST server-client systems some python values cannot be decoded into json and hence cannot be used. Json ist used to send the brain file via the REST interface to NEST server.

 

To:

  1. Only midsize brains can be run on Castor anyway, since very large brains will eat up all memory on Castor VMs. I suggest to default to PizDaint for larger brain models, which worked for this experiment.

  2. Is a native property of the REST interface and shall be addressed by the NEST team.

  3. Must be soved by the EBRAINS OIDC team. Currently the issue can be circumvented by the user by logging in to the Virtual Coach before any VC command in the users’s script. Also compare this issue to a similar one:

·

Viktor Vorobev 
December 16, 2022 at 1:46 PM

Hi Felix! Could you tell, please, if the NRP versions for the VC and the frontend usage are the same? Do you use the same instance of the NRP, of local vs online (http://148.187.149.212/#/esv-private)?

·

Anonymous 
December 15, 2022 at 2:29 PM

Any update on this ticket?

·

Details

Assignee

Priority

Created October 18, 2022 at 8:11 AM
Updated December 16, 2023 at 8:49 PM