Draft cluster snakefile stuff.

The "Snakefile_cluster_iteration" file manages the list of
widths, and can iterate over jobs on the cluster and then
do a (trivial) aggregation step.

The queuing_config.yaml file is a short cluster config that
removes the need to type the cluster stuff into the command
line.  It conventionally goes in
$HOME/.config/snakemake/queuing/config.yaml, and is accessed
by "snakemake <other-args> --profile=queuing".

The Snakefile has mpi stuff for the trivial aggregation rule
because the cluster-ness of it applies to all the rules.
This is clumsy, but not wrong.

episodes/files/Snakefile_cluster_iteration

@@ -0,0 +1,24 @@
+#
+# Run a bunch of Amdahl jobs and aggregate the output.
+#
+WIDTHS=[1,2]
+#
+def getwidth(wildcards):
+    return wildcards.sample
+
+rule plot:
+  input: expand('{size}.out',size=WIDTHS)
+  output: 'done.out'
+  resources:
+    mpi="mpirun",
+    tasks=1
+  shell: 'echo "{WIDTHS}, Done!" > done.out'
+rule iterate:
+  input: 
+  output: '{sample}.out' 
+  resources:
+    mpi="mpirun",
+    tasks=getwidth
+  shell: 
+    "module load OpenMPI; mpirun -np {resources.tasks} amdahl > {wildcards.sample}.out"
+

episodes/files/queuing_config.yaml

@@ -0,0 +1,6 @@
+#  snakemake -j 3 --cluster "sbatch -N 1 -n {resources.tasks} -p node"
+cluster: 
+  sbatch
+    --partition=node
+    --nodes=1
+    --tasks={resources.tasks}