Fix merging responses in nest-server-mpi
#3492
Conversation
babsey
force-pushed
the
nest-server-mpi-merge-responses
branch
from
78404ae to
909e5eb
Compare
babsey
changed the title
nest-server-mpi
gtrensch
added
T: Bug
heplesser
added
S: High
steffengraber
left a comment
steffengraber
left a comment
There was a problem hiding this comment.
+1 The logic seems perfectly correct to me. Robustness could be increased by additional checks of values and lists. But only if necessary.
|
I am working on better solution of combine function. Thank you for the patience. |
heplesser
left a comment
heplesser
left a comment
There was a problem hiding this comment.
Overall this looks good to me, but I think docstrings for the new methods would be an advantage for long-term maintenance of the code.
…est-simulator into nest-server-mpi-merge-responses
pynest/nest/server/hl_api_server.py
Outdated
| * the responses are known to be the same from the master and all | ||
| workers. In this case, the combined response is just the master | ||
| response. |
There was a problem hiding this comment.
Please avoid loaded terms such as "master" and "workers". All MPI ranks in a NEST simulation are doing the same work, just on different parts of the network. Also, how would it be "known" that we get the same response from each rank?
There was a problem hiding this comment.
As I mentioned, it was written mainly on @jougs effort. I am not able to join into the discussion about MPI concept which I am not fully understand of.
pynest/nest/server/hl_api_server.py
Outdated
| * if the response contains one list per process, the combined | ||
| response will be those first list. |
There was a problem hiding this comment.
I don't understand this, especially the "those first list". If I get one list from each rank and the data fall not under the "same from all" clause above, don't I need to combine the lists from all ranks?
|
I apogize for the bad PR. I am not well qualified to resolve this. I suggest that it would be part of the hackathon topic. |
Co-authored-by: Hans Ekkehard Plesser <hans.ekkehard.plesser@nmbu.no>
|
Pull request automatically marked stale! |
github-actions
bot
added
the
stale
4 participants
When executing simulation via
nest-server-mpi, the result seems not correctly.In fact, when using a number of processes greater than 1, lets say an example
np = 3.It shows only activity of each 3rd neuron.
It seems that we need to merge the responses of
nest-server-mpi.With this PR, NEST Desktop shows activity of all neurons executed on
nest-server-mpi.