-
Notifications
You must be signed in to change notification settings - Fork 25
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
experiment class for laghos #362
base: develop
Are you sure you want to change the base?
Conversation
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Nice job so far! Some updates requested, let me know if you have any questions.
repo/laghos/application.py
Outdated
workload_variable('rp', default='0', | ||
description='number of parallel refinements', | ||
workloads=['laghos']) | ||
workload_variable('ms', default='4', |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Likewise let's update this default value back to 500
def compute_applications_section(self): | ||
variables = {} | ||
|
||
if self.spec.satisfies("scaling=weak"): |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Let's start with the scaling assuming a single node, but change problem size per rank instead, once we know we have this right we can add more nodes or make updates. So for weak scaling:
rs values: 1
num processes: 8
rp values (increases workload per rank): 1,2,3,4
Another option if you can also implement this is rs values 1,2,3 and mpi processes 1,8,64
variables["rs"] = ["5","6"] | ||
variables["n_nodes"] = ["1","8"] | ||
elif self.spec.satisfies("scaling=strong"): | ||
variables["n_nodes"] = ["1","8"] |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Instead of increasing nodes for strong scaling let's keep it on one node and increase MPI ranks, 1,8,64. We should lower the rs value to 1 or 2 here as well, depending on how long this takes to solve, these would ideally be less than 5 minutes to solve.
"problem": { | ||
#"variables": variables, | ||
"experiments": { | ||
"laghos_{n_nodes}_{n_ranks}": { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Can you add the other parameters into the naming scheme here, so basically anything that gets passed in (rs, rp, etc) should be in here, see the naming for the kripke experiment in https://github.com/LLNL/benchpark/pull/344/files
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@august-knox I enabled a dry-run test and it is failing. Please diagnose/fix.
@@ -0,0 +1,74 @@ | |||
from benchpark.directives import variant |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@pearce8 @august-knox We need to remove this file. I just kept it around for comparison
remove old format
…into experiment/laghos
removing old laghos location
…into experiment/laghos
removing smb
removing smb
…into experiment/laghos
bin/benchpark experiment init --dest laghos
Don't believe there's any variants that need to be specified here