Energy consumption


#1

Hello!

I am trying to monitor the energy consumption of an SNN implemented in SpiNNaker. I have seen that there is a way to receive this information in the ‘reports’ file that can be downloaded after finishing a job by setting “reports_enabled = True” and “write_energy_report = True” in the cfg file as said here: https://github.com/SpiNNakerManchester/sPyNNaker/issues/318

However, I don’t know how to access this file using the neuromorphic computing platform. If anyone has come up with a solution to monitoring energy consumption in this way or another, I would appreciate the help.

Thanks in advance
Rafael Pérez Belizón


#2

Dear Rafael,

Although I have already replied to you directly via email, I will also do the same on this forum so that the information is available.

Regarding the use of SpiNNaker via the Collaboratory, I haven’t yet found, if possible, a way to access the .cfg file.

The alternative solution that I propose is to use the direct access to SpiNNaker with JupyterLab (https://spinn-20.cs.man.ac.uk). Indeed, using the terminal, a .spynnaker.cfg file is in your home directory. By modifying it as suggested in the github issue above and after running a simulation, the reports seem accessible in the ~/reports directory. There are two report files that may interest you in the subdirectory corresponding to your finished job:
Detailed_energy_report.rpt
energy_summary_report.rpt
Note that JupyterLab access is accessible with your EBRAINS credentials.

Best regards,

Matthieu


#3

Hi Rafael,

The way I have done this in the past is by creating a zip with my script file(s) and a config file in the same directory as the script you call to run the job. This config file needs to be called spynnaker.cfg, and in it, in order to get the energy report you need to have the following text:

[Reports]
reports_enabled = True
write_energy_report = True

To run this on the neuromorphic computing platform, put the zip somewhere accessible online and then choose the “From Git repository or zip archive” option and put the http link to the zip file in there. Then in the “command” box simply put the full name of the script file (i.e. script.py) and any command line arguments if you have them, then you should be able to click on submit and your job will run.

Once your job has finished running successfuly, you should then be able to find the relevant energy reports in the reports.zip file.

Having said all this, at the moment we have discovered there is a bug that means the spynnaker.cfg file does not currently get parsed correctly from the zip and is therefore ignored. You may therefore be better for now in following Matthieu’s suggestion and using our JupyterLab setup at the address in his message.

Andrew Gait


#4

Hi Rafael,

Just to add further to this, we have now fixed the bug and you should now be able to run using a cfg file on the Collaboratory following the instructions in my previous reply; in addition, to get the energy report, you will need to run using the master branch of the tools by putting

{
“spynnaker_version”: “master”
}

into the hardware configuration box.

Andrew


#5

Many thanks for this help!