🚀 go-pugleaf

I'm a new, non-administartor type user of AIX & xntpd on a cluster of 32
RS6000's.  This system is used for lots of compute-bound things, some running
in parallel over Ethernet.

Anyway, the nodes don't seem to be keeping very good time:

	lace01 13:58:18
	lace02 13:58:21 +3
	lace03 13:58:20 -1
	lace04 13:58:21 +1
	lace05 13:58:23 +2
	lace06 13:58:24	+1
	lace07 13:58:35 +11
	lace08 13:58:26 -9
	lace09 13:58:29 +3
	lace10 13:58:28 -1
	lace11 13:58:29 +1
	lace12 13:58:32 +3
	lace13 13:58:37 +5
	lace14 13:58:31 -6
	lace15 13:58:34 +3
	lace16 13:58:34
	lace17 13:58:35 +1
	lace18 13:58:32 -3
	lace19 13:58:39 +7
	lace20 13:58:39
	lace21 13:58:41 +2
	lace22 13:58:39 -2
	lace23 13:58:42 +3
	lace24 13:58:40 -2
	lace25 13:58:46 +6
	lace26 13:58:47 +1
	lace27 13:58:43 -4
	lace28 13:58:46 +3
	lace29 13:58:44 -2
	lace30 13:58:47 +3
	lace31 13:58:53 +6
	lace32 13:58:47 -6

The above output is from a trivial rsh script run from lace07, the elapsed time
was 29 seconds.

My question is, are the above time descrepancies reasonable?  For instance,
there's a 9 second reverse jump between lace08 and lace07.  From listening
to this group for a little while it sounds like things should be better.
On the other hand, with bursty network traffic and often compute-bound CPUs
I could imagine xntpd going a bit 'unstable' attempting to track the clock
drifts while seeing varying network delays.

I'd appreciate any comments you may have.  I`m not an administrator for this
system but if the net wisdom can suggest some things to try, either I can
try it or if I don't have the required permissions I think I can find someone
who does.



-- 
Scott Townsend,  Sverdrup Technology Inc.  NASA Lewis Research Center Group
fsset@bach.lerc.nasa.gov