Igor,

I believe that the NTP protocol recommends 3 time servers as part of the algorithm calls for checking for network latency, and keeping in mind that the "authority" time server may drift as well.

If you think that your 740 is "more stable", add that to the list of time servers (I think that the filer will allow a query), then it will get even more interesting. I know that on unix servers, it is suggested that localhost be added as a low priority host - that way, just in case the network bogs down, it can rely on the local clock without having to reset it all the time.

Hope this helps,

Jim

-----Original Message----- From: Igor Schein [mailto:igor@txc.com] Sent: Friday, April 12, 2002 4:00 PM To: toasters@mathworks.com Subject: time sync issue

Hi,

I have 2 filers, 740 running 6.1.1R2 and 840 running 6.2R1.

I have the same timed setup on both machines:

timed.enable on timed.log on timed.max_skew 30m timed.min_skew 50 timed.proto ntp timed.sched hourly timed.servers ntphost1

Here's how the messages look for 740 filer:

Time reported by time server ntphost1 is Fri Apr 12 13:10:54 GMT 2002 (skew +0.088 secs). Time reported by time server ntphost1 is Fri Apr 12 14:08:34 GMT 2002 (skew +0.055 secs). Time reported by time server ntphost1 is Fri Apr 12 14:57:16 GMT 2002 (skew +0.129 secs). Time reported by time server ntphost1 is Fri Apr 12 17:11:09 GMT 2002 (skew +0.051 secs). Time reported by time server ntphost1 is Fri Apr 12 18:01:03 GMT 2002 (skew +0.058 secs). Time reported by time server ntphost1 is Fri Apr 12 19:01:11 GMT 2002 (skew +0.116 secs). Time reported by time server ntphost1 is Fri Apr 12 19:55:54 GMT 2002 (skew +0.080 secs). Time reported by time server ntphost1 is Fri Apr 12 20:06:43 GMT 2002 (skew +0.054 secs). Time reported by time server ntphost1 is Fri Apr 12 20:37:57 GMT 2002 (skew +0.098 secs). Time reported by time server ntphost1 is Fri Apr 12 21:05:39 GMT 2002 (skew +0.102 secs).

The skew is always positive and always small ( 0.1 secs or less ).

Now here're the messages for 840 filer:

Time reported by time server ntphost1 is Fri Apr 12 12:05:10 GMT 2002 (skew -0.177 secs). Time reported by time server ntphost1 is Fri Apr 12 13:05:32 GMT 2002 (skew -0.221 secs). Time reported by time server ntphost1 is Fri Apr 12 13:52:17 GMT 2002 (skew +0.238 secs). Time reported by time server ntphost1 is Fri Apr 12 14:03:44 GMT 2002 (skew -0.065 secs). Time reported by time server ntphost1 is Fri Apr 12 16:08:06 GMT 2002 (skew -0.101 secs). Time reported by time server ntphost1 is Fri Apr 12 17:00:57 GMT 2002 (skew -0.102 secs). Time reported by time server ntphost1 is Fri Apr 12 18:06:15 GMT 2002 (skew -0.199 secs). Time reported by time server ntphost1 is Fri Apr 12 18:53:44 GMT 2002 (skew +0.185 secs). Time reported by time server ntphost1 is Fri Apr 12 19:08:31 GMT 2002 (skew -0.243 secs). Time reported by time server ntphost1 is Fri Apr 12 20:01:58 GMT 2002 (skew -0.140 secs).

The skew's sign jumps back and forth, and the absolute value is over 0.2 secs ( becomes as high as 0.3 secs, which prompts make utility on clients to complain ).

What can be wrong with the clock on 840 filer? I suppose I can change timed.sched to sync more frequently, but I am just curious about the nature of the issue.

Thanks

Igor