[SOLVED] Nengo and Spinnaker


#1

Hello everyone,

I am trying to use Nengo on the HB colab with Spinnaker, there is one topic about the integration of Nengo but it is old so I was wondering if there is any advancement on that.

I tried to add Nengo and all the different needed libraries by using pip, and in order to make it work using Nengo_Spinnaker, I need also to add some information about the Spinnaker board (IP address etc) like in this example.

Is it possible to get such information for the configuration? If not, is there any possible way to make it work?

Thank you.


#2

Hi,

I think the issue is that Nengo_SpiNNaker doesn’t support the SpiNNaker allocation service “spalloc”. The workaround for this on Jupyter would be to go to the “terminal” and run spalloc from there. Once you have a terminal, this would be done as follows:

cd sPyNNaker
source bin/activate
spalloc -H spinnaker.cs.man.ac.uk --owner <email address> <n_boards>

where:

  • email_address is your email address
  • n_boards is the number of boards you need

This will then give you the IP address of the boards. The command will continue to run until you press “enter”, and this will keep the boards until then. Note that if you logout though, the allocation will be lost.

Let me know if you have any trouble with this.

Andrew :slight_smile:


#3

Thank you @rowley ! it is working now. I tried a simple example and I can see a similar output with/without the use of Spinnaker, however it is taking more time to execute in Spinnaker. Is that because of the boot time of the reserved board?


#4

Hi,

I don’t have a lot of experience of this but that is possible, as the boot and configuration can take a bit of time. I believe that the solution works best when it is run on larger examples where the benefits of the parallelism of SpiNNaker then become more apparent. That said, it will depend on the dimensionality of the problem, as at quite large scale, the communications become too much for the machine to cope with as the code is currently written.

Andrew :slight_smile:


#5

Ok, thank you very much for the explanation.