[updated] Incorporate Cajita / Cabana::Grid load balancer #113
Conversation
|
@streeve @aetx This is an updated version of PR #104 (see PR description for an overview of changes). The CI is pending approval, but based on the runs in my fork (cz4rs#1), all builds are passing. For the context, we are exploring using CabanaMD as a benchmark for load balancing with communication. |
|
If I remember correctly, the only ugly hack was in I don't remember another reason for it being not merged. |
cz4rs
force-pushed
the
cabana_lb
branch
2 times, most recently
from
ab89b0d to
5ab91d5
Compare
|
Sorry I lost track of this PR and thanks very much for pushing this forward. Would you be okay with merging your general (and very helpful) changes in one PR, then merging the load balancing in a second? |
|
Sure, I will split the changes.
Please note that I'm on vacation now, so this will need to wait a bit (I'll
be back on September 9th).
pt., 30 sie 2024, 16:12 użytkownik Sam Reeve ***@***.***>
napisał:
… Sorry I lost track of this PR and thanks very much for pushing this
forward. Would you be okay with merging your general (and very helpful)
changes in one PR, then merging the load balancing in a second?
—
Reply to this email directly, view it on GitHub
<#113 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/ABDMTPPUTPIEXHUC6CBM473ZUB4ULAVCNFSM6AAAAABJZQHOIWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGMRRGM4TEOBUGY>
.
You are receiving this because you authored the thread.Message ID:
***@***.***>
|
|
@streeve Please have a look. I have some more changes (extending LAMMPS commands support) in DARMA-tasking#3, I will post those in a separate PR. |
|
@streeve Ping |
src/inputFile_impl.h
Outdated
| @@ -455,7 +512,14 @@ void InputFile<t_System>::create_lattice( Comm<t_System> *comm ) | |||
| T_X_FLOAT max_z = lattice_constant * lattice_nz; | |||
| std::array<T_X_FLOAT, 3> global_low = { 0.0, 0.0, 0.0 }; | |||
| std::array<T_X_FLOAT, 3> global_high = { max_x, max_y, max_z }; | |||
| system->create_domain( global_low, global_high ); | |||
| if ( commandline.sparse ) | |||
There was a problem hiding this comment.
Since this commandline arg is already set up, can we input the multiplier here instead of assuming 2?
There was a problem hiding this comment.
Sure, I have missed this somehow 🤦♂️
There was a problem hiding this comment.
I think that was the original behavior to get results, but this will be much more general
There was a problem hiding this comment.
Done. Example below with 1.5 multiplier (mpiexec -n 8 ./build/bin/cbnMD -il input/in.lj --vacuum 1.5).
Place rank before timestep and add leading zeroes in the filenames. This helps ParaView deduce the order correctly.
cz4rs
force-pushed
the
cabana_lb
branch
2 times, most recently
from
05ccc9c to
06c5a50
Compare
Updated version of #104. I haven't touched the logic of it all, just a rebase, clean-up and dropping some redundant changes.
--sparsecommand line flag for an unbalanced system to avoid having to change the source