Index: trunk/Ohana/src/opihi/doc/pantasks.txt
===================================================================
--- trunk/Ohana/src/opihi/doc/pantasks.txt	(revision 8190)
+++ trunk/Ohana/src/opihi/doc/pantasks.txt	(revision 11898)
@@ -1,2 +1,27 @@
+
+- task spawning speed
+
+  I have been examining things which affect the speed of the pantasks
+  processing. I have learned some interesting things:
+
+  * pcontrol was being slammed with requests for status by pantasks.
+    this may account for Paul's controller hang-ups.  I have added a
+    long (500ms) sleep to the controller thread to limit the rate at
+    which controller checks are run
+
+  * adding even a small usleep to the task_thread or job_thread puts
+    them to sleep for a long time (>> 10ms).  it seems longer than the
+    linux time slicer.  I have removed sleeps from the task and job
+    threads.
+
+  * the job submit rate is apparently limited by two things:
+
+    * when the job is submitted (SubmitJob) the interaction with
+      the controller seems to take ~30ms or more.
+
+    * some thread (controller thread? main readline thread?) seems to
+      introduce timeouts which are very long (up to 100ms).  These
+      introduce bit delays if when they happen during the task_thread
+      loop. 
 
 - updates for queues: