Index: /trunk/Ohana/doc/www/html/IPP-subsystems/pcontrol/speedtest.hts
===================================================================
--- /trunk/Ohana/doc/www/html/IPP-subsystems/pcontrol/speedtest.hts	(revision 6710)
+++ /trunk/Ohana/doc/www/html/IPP-subsystems/pcontrol/speedtest.hts	(revision 6710)
@@ -0,0 +1,42 @@
+
+I ran 8 x 32 fake image copies on the 32 po nodes. each one should
+last a minimal amount of time (no real action, no sleep).  The time to
+process the stack is essentially the speed at which pcontrol can cycle
+though the nodes:
+
+-rw-rw-r--  1 eugene users   17 2005-08-11 19:28:50.000000000 -1000 file.00.032.fits
+-rw-rw-r--  1 eugene users   17 2005-08-11 19:29:20.000000000 -1000 file.07.015.fits
+
+that is 30 seconds to run 8 x 32 jobs: 
+
+0.117 sec / job
+3.750 sec / (job / node)
+
+I ran the same 8 x 32 jobs on one node, with the following rate:
+
+-rw-rw-r--  1 eugene users   17 2005-08-11 19:34:39.000000000 -1000 file.00.001.fits
+-rw-rw-r--  1 eugene users   17 2005-08-11 19:35:30.000000000 -1000 file.07.029.fits
+
+or 51 seconds to run 8 x 32 jobs on 1 node.
+
+0.199 sec / job
+0.199 sec / (job / node)
+
+--
+
+I ran the same test getting 56 seconds for 8 x 32 copies on 32 nodes.
+This time, I used one of the machines in the po cluster (po33) for the
+test:
+
+0.219 sec / job
+
+-rw-rw-r--  1 eugene users   17 2005-08-11 19:45:25.000000000 -1000 file.00.030.fits
+-rw-rw-r--  1 eugene users   17 2005-08-11 19:46:21.000000000 -1000 file.07.028.fits
+
+--
+
+running on psched, from kiawe to po01-po32
+-rw-rw-r--  1 eugene users    17 2005-08-11 20:11:25.000000000 -1000 file.00.003.fits
+-rw-rw-r--  1 eugene users    17 2005-08-11 20:12:30.000000000 -1000 file.07.032.fits
+
+55 seconds; roughly equivalent.  
Index: /trunk/Ohana/doc/www/html/IPP-subsystems/psched/speedtests.hts
===================================================================
--- /trunk/Ohana/doc/www/html/IPP-subsystems/psched/speedtests.hts	(revision 6710)
+++ /trunk/Ohana/doc/www/html/IPP-subsystems/psched/speedtests.hts	(revision 6710)
@@ -0,0 +1,15 @@
+
+CheckController is critical for speed
+
+parse jobs: 1.5 usec per job + 20 usec
+
+check exit stack: 2.7ms average, up to 10ms: this seems to be the
+kernel scheduler putting pcontrol in the background every now and
+again.  
+
+with 1024 jobs and 4 machines, the jobs being sent to machines (should
+take only 1 second, but timeouts add in delays) takes up the first few
+seconds.  initially
+
+
+*write this up, add some tests for timing... *