debbieg@smokey.sandiego.NCR.COM (03/23/91)
I am running xntpd on NCR System 3000's utilizing UNIX System V.4. In
my current configuration I have 40-50 stratum 4 systems which
synchronize to the same 3 stratum 3 servers. With this setup, the
servers drop many packets causing them to become unreachable to the
stratum 4 systems. See as follows:
server1 server2 server3
xntpdc> iostats
time since reset: 17740 17044 16512
total receive buffers: 25 30 20
free receive buffers: 24 29 19
used receive buffers: 0 0 0
low water refills: 3 4 2
dropped packets: 5892 5186 6412
ignored packets: 0 0 0
received packets: 3191 3973 2494
packets sent: 3817 4125 2756
packets not sent: 0 0 0
interrupts handled: 3131 3803 2446
received by interrupt: 9083 9159 8906
The stratum 4 systems poll every 64 seconds when a server is
reachable.
Are 40-50 pollers too many for a server to process?
Is the poll value of 64 seconds to small?
Is my problem elsewhere?
Thanks for any assistance!
Debbie.Galeazzi@SanDiego.NCR.COM
NCR Corporation
E&M - San Diego, CaliforniaMills@udel.edu (03/23/91)
Debbie, 40-50 chimers per server does not sound like a significant load. I see something like 200 now per fuzzball primary server and they seen to grunt along without dropping many packets, usually none. However, if you have changed the min- poll value or max-poll value, you have defeated the automatic poll interval calculation, which starts out at 64 seconds and then ratchets back to 1024 seconds as conditions permit. In NTP Version 2, which you are running now, the poll interval is clamped to 64 seconds for the peer(s) a client is actually synchronized to; while in NTP Version 3, even those peers ratchet up as well. You may address your cards and letters to the implementors of a v3 daemon, should they care to identify themselves. Dave