power saving SLURM script

README.md

@@ -169,6 +169,43 @@ It may be useful to remember the sinfo codes:
 \^  The node reboot was issued.  
 \-  The node is planned by the backfill scheduler for a higher priority job.  
 
+#### slurm.conf script
+```
+#!/bin/bash
+# PowerSaving Segment
+# How long shall we wait for a node to be idle before shutting it down?
+SuspendTime=300 # 5 minutes
+# How many nodes will your utility let you stop at once?
+SuspendRate=40
+# How many nodes will your utility let you start at once?
+ResumeRate=40
+# Where is the script to shut the nodes down?
+SuspendProgram=/opt/system/slurm/etc/suspend.sh
+# Where is the script to start the nodes up?
+ResumeProgram=/opt/system/slurm/etc/resume.sh
+# Time how long does it take for your slowest node to shutdown? Add 60 to that and put the answer below.
+SuspendTimeout=300
+# Time how long it takes for your slowest node to start up? Add 60 to that and put the answer below.
+ResumeTimeout=240
+# If you want to exclude certain nodes or partitions, enable the below and fill them in.
+#SuspendExcNodes=compute-3
+#SuspendExcParts=main
+# How long to wait between a node being up, and when a job runs. This can take awhile, FYI.
+BatchStartTimeout=360 # Default is 10.
+# Time for messages to go from the management node and a compute node AND back. This can take awhile.
+MessageTimeout=100 # Default is 10.
+
+# I kept the partition and node information below the above.
+...
+
+# At the very bottom I changed/added the below. The first is to prevent the nodes from being 'down' when they suspend.
+# Which requires some scontrol commands to return them to service.
+SlurmctldParameters=enable_configless,idle_on_node_suspend
+# This one returns a node to service when it registers itself as up.
+ReturnToService=2
+```
+
+