I've just had several runs crash on pitagora with the following error written to standard out:
"
There are 15 shared-memory domains. This domain has 48 CPUs.
Error: The number of CPUs needs to be divisible by the number of shared memory domains. Aborting
"
Here's the parallelisation setup in case helpful:
Total number of grid points:
nxnynznmunvpanspec = 467140608 = 445.50 MiB = .44 GiB.
nkxnkynznmunvpanspec = 106216704 = 101.30 MiB = .10 GiB.
nkxnkynz = 184404 = .18 MiB = .00 GiB.
Number of points to be parallelised:
vmu-layout: 1152 (nmunvpanspec)
kxkyz-layout: 184404 (nkxnkynzntubesnspec)
Number of points per processor:
vmu-layout: 1
kxkyz-layout: 161
I have not selected any special option for shared-memory, so I think it is the default. I don't understand how the number of shared-memory domains is determined, but it does not seem like ideal behaviour for runs to fail for this reason. As things stand, I have no idea how many cores I am allowed to request for a given simulation. Is there a way to overcome this?
I've just had several runs crash on pitagora with the following error written to standard out:
"
There are 15 shared-memory domains. This domain has 48 CPUs.
Error: The number of CPUs needs to be divisible by the number of shared memory domains. Aborting
"
Here's the parallelisation setup in case helpful:
Total number of grid points:
nxnynznmunvpanspec = 467140608 = 445.50 MiB = .44 GiB.
nkxnkynznmunvpanspec = 106216704 = 101.30 MiB = .10 GiB.
nkxnkynz = 184404 = .18 MiB = .00 GiB.
Number of points to be parallelised:
vmu-layout: 1152 (nmunvpanspec)
kxkyz-layout: 184404 (nkxnkynzntubesnspec)
Number of points per processor:
vmu-layout: 1
kxkyz-layout: 161
I have not selected any special option for shared-memory, so I think it is the default. I don't understand how the number of shared-memory domains is determined, but it does not seem like ideal behaviour for runs to fail for this reason. As things stand, I have no idea how many cores I am allowed to request for a given simulation. Is there a way to overcome this?