allow for partial access to huge & largemem nodes (#234)

OSC · Feb 24, 2023 · f9b7138 · f9b7138
1 parent 01603cf
commit f9b7138
Show file tree

Hide file tree

Showing 6 changed files with 144 additions and 32 deletions.
diff --git a/apps.awesim.org/apps/bc_desktop/owens.yml.erb b/apps.awesim.org/apps/bc_desktop/owens.yml.erb
@@ -82,8 +82,24 @@ attributes:
  - **hugemem** - (*48 cores*) This Owens node has 1.5TB of available RAM
  as well as 48 cores. There are 16 of these nodes on Owens.
  options:
- - [ "any", "any" ]
- - [ "vis", "vis" ]
- - [ "gpu", "gpu-any" ]
- - [ "hugemem", "hugemem" ]
+ - [
+ "any", "any",
+ data-min-cores: 1,
+ data-max-cores: 28,
+ ]
+ - [
+ "vis", "vis",
+ data-min-cores: 1,
+ data-max-cores: 28,
+ ]
+ - [
+ "gpu", "gpu-any",
+ data-min-cores: 1,
+ data-max-cores: 28,
+ ]
+ - [
+ "hugemem", "hugemem",
+ data-min-cores: 4,
+ data-max-cores: 48,
+ ]
 submit: submit/slurm.yml.erb
diff --git a/apps.awesim.org/apps/bc_desktop/pitzer.yml.erb b/apps.awesim.org/apps/bc_desktop/pitzer.yml.erb
@@ -87,12 +87,54 @@ attributes:
  [detailed information on the Pitzer cluster]: https://www.osc.edu/resources/technical_support/supercomputers/pitzer
  [NVIDIA Tesla V100 GPUs]: https://www.nvidia.com/en-us/data-center/v100/
  options:
- - [ "any", "any" ]
- - [ "48 core", "any-48core" ]
- - [ "any gpu", "gpu-any"]
- - [ "48 core with gpu", "gpu-48core" ]
- - [ "densegpu", "densegpu" ]
- - [ "visualization node", "vis"]
- - [ "largemem", "largemem" ]
- - [ "hugemem", "hugemem" ]
+ - [
+ "any", "any",
+ data-min-cores: 1,
+ data-max-cores: 80,
+ ]
+ - [
+ "40 core", "any-40core",
+ data-min-cores: 1,
+ data-max-cores: 40,
+ ]
+ - [
+ "48 core", "any-48core",
+ data-min-cores: 1,
+ data-max-cores: 48,
+ ]
+ - [
+ "any gpu", "gpu-any",
+ data-min-cores: 1,
+ data-max-cores: 48,
+ ]
+ - [
+ "40 core with gpu", "gpu-40core",
+ data-min-cores: 1,
+ data-max-cores: 40,
+ ]
+ - [
+ "48 core with gpu", "gpu-48core",
+ data-min-cores: 1,
+ data-max-cores: 48,
+ ]
+ - [
+ "densegpu", "densegpu",
+ data-min-cores: 1,
+ data-max-cores: 48,
+ ]
+ - [
+ "visualization node", "vis",
+ data-min-cores: 1,
+ data-max-cores: 48,
+ ]
+ - [
+ "largemem", "largemem",
+ data-min-cores: 24,
+ data-max-cores: 48,
+ ]
+ - [
+ "hugemem", "hugemem",
+ data-min-cores: 20,
+ data-max-cores: 80,
+ ]
 submit: submit/slurm.yml.erb
diff --git a/apps.awesim.org/apps/bc_desktop/submit/slurm.yml.erb b/apps.awesim.org/apps/bc_desktop/submit/slurm.yml.erb
@@ -31,8 +31,6 @@
  end
  end
 
- hugemem_cpus = cluster == 'pitzer' ? 80 : 48
-
  slurm_args = case node_type
  # 'any' case handled by scheduler, this is just a quick short circuit
  when "any"
@@ -57,10 +55,10 @@
  # using partitions here is easier than specifying memory requests
  when "largemem"
  partition = bc_num_slots.to_i > 1 ? "largemem-parallel" : "largemem"
- base_slurm_args + p20_node + ["--partition", partition ]
+ base_slurm_args + tasks_per_node + ["--partition", partition ]
  when "hugemem"
  partition = bc_num_slots.to_i > 1 ? "hugemem-parallel" : "hugemem"
- base_slurm_args + [ "--ntasks-per-node", "#{hugemem_cpus}", "--partition", partition ]
+ base_slurm_args + tasks_per_node + ["--partition", partition ]
 
  else
  base_slurm_args

diff --git a/ondemand.osc.edu/apps/bc_desktop/owens.yml.erb b/ondemand.osc.edu/apps/bc_desktop/owens.yml.erb
@@ -82,8 +82,24 @@ attributes:
  - **hugemem** - (*48 cores*) This Owens node has 1.5TB of available RAM
  as well as 48 cores. There are 16 of these nodes on Owens.
  options:
- - [ "any", "any" ]
- - [ "vis", "vis" ]
- - [ "gpu", "gpu-any" ]
- - [ "hugemem", "hugemem" ]
+ - [
+ "any", "any",
+ data-min-cores: 1,
+ data-max-cores: 28,
+ ]
+ - [
+ "vis", "vis",
+ data-min-cores: 1,
+ data-max-cores: 28,
+ ]
+ - [
+ "gpu", "gpu-any",
+ data-min-cores: 1,
+ data-max-cores: 28,
+ ]
+ - [
+ "hugemem", "hugemem",
+ data-min-cores: 4,
+ data-max-cores: 48,
+ ]
 submit: submit/slurm.yml.erb
diff --git a/ondemand.osc.edu/apps/bc_desktop/pitzer.yml.erb b/ondemand.osc.edu/apps/bc_desktop/pitzer.yml.erb
@@ -87,12 +87,54 @@ attributes:
  [detailed information on the Pitzer cluster]: https://www.osc.edu/resources/technical_support/supercomputers/pitzer
  [NVIDIA Tesla V100 GPUs]: https://www.nvidia.com/en-us/data-center/v100/
  options:
- - [ "any", "any" ]
- - [ "48 core", "any-48core" ]
- - [ "any gpu", "gpu-any"]
- - [ "48 core with gpu", "gpu-48core" ]
- - [ "densegpu", "densegpu" ]
- - [ "visualization node", "vis"]
- - [ "largemem", "largemem" ]
- - [ "hugemem", "hugemem" ]
+ - [
+ "any", "any",
+ data-min-cores: 1,
+ data-max-cores: 80,
+ ]
+ - [
+ "40 core", "any-40core",
+ data-min-cores: 1,
+ data-max-cores: 40,
+ ]
+ - [
+ "48 core", "any-48core",
+ data-min-cores: 1,
+ data-max-cores: 48,
+ ]
+ - [
+ "any gpu", "gpu-any",
+ data-min-cores: 1,
+ data-max-cores: 48,
+ ]
+ - [
+ "40 core with gpu", "gpu-40core",
+ data-min-cores: 1,
+ data-max-cores: 40,
+ ]
+ - [
+ "48 core with gpu", "gpu-48core",
+ data-min-cores: 1,
+ data-max-cores: 48,
+ ]
+ - [
+ "densegpu", "densegpu",
+ data-min-cores: 1,
+ data-max-cores: 48,
+ ]
+ - [
+ "visualization node", "vis",
+ data-min-cores: 1,
+ data-max-cores: 48,
+ ]
+ - [
+ "largemem", "largemem",
+ data-min-cores: 24,
+ data-max-cores: 48,
+ ]
+ - [
+ "hugemem", "hugemem",
+ data-min-cores: 20,
+ data-max-cores: 80,
+ ]
 submit: submit/slurm.yml.erb
diff --git a/ondemand.osc.edu/apps/bc_desktop/submit/slurm.yml.erb b/ondemand.osc.edu/apps/bc_desktop/submit/slurm.yml.erb
@@ -31,8 +31,6 @@
  end
  end
 
- hugemem_cpus = cluster == 'pitzer' ? 80 : 48
-
  slurm_args = case node_type
  # 'any' case handled by scheduler, this is just a quick short circuit
  when "any"
@@ -57,10 +55,10 @@
  # using partitions here is easier than specifying memory requests
  when "largemem"
  partition = bc_num_slots.to_i > 1 ? "largemem-parallel" : "largemem"
- base_slurm_args + p20_node + ["--partition", partition ]
+ base_slurm_args + tasks_per_node + ["--partition", partition ]
  when "hugemem"
  partition = bc_num_slots.to_i > 1 ? "hugemem-parallel" : "hugemem"
- base_slurm_args + [ "--ntasks-per-node", "#{hugemem_cpus}", "--partition", partition ]
+ base_slurm_args + tasks_per_node + ["--partition", partition ]
 
  else
  base_slurm_args