nanog mailing list archives

Re: Interesting stratum 1 NTP clock


From: Jerry Scharf <scharf () vix com>
Date: Wed, 24 Jun 1998 11:08:11 -0700


Would a GPS hooked to the serial port of a Linux box do the same?

Dirk

On Mon, Jun 22, 1998 at 10:43:48PM -0700, Tony Li wrote:

Folks who run NTP in their nets might wanna check out:

http://www.coetanian.com/tss/tss100.htm

This is a reasonably affordable, NTP-ready, GPS based stratum 1 chimer,
complete with 10BaseT interface.  It's also getting close to being
plug-n-play.

I've been beta testing one for a couple of weeks and it seems to chime
reasonably well.  

Tony

p.s. I have no finanical interest in this product or this company
whatsoever.  I'm just a time geek...

Any clock hooked up to a NTP server will make the server a stratum 1 server. 
It makes no statements about accuracy, just NTP hopwise closeness to truth. To 
NTP stratum, there is no difference between the thing Tony mentioned and a 
modem calling ACTS. Your accuracy of time will be much lower in the later 
cases, with what you proposed usually having errors in the single digit 
milliseconds.

(rant #2)

Lots of people say 'Why should I care about accurate time?' I have used NTP 
synchronized clock with both tcpdump and Cisco logging to chase problems. 
Sometimes I'm after big things, but other times I'm trying to figure out 
whether two things that are happening are the same or different. Being able to 
have time accurate to the few packet range is a real in separating these out. 
It's really hard to get there when you're talking about OC48s, but even fast 
ethernets have packets lasting in the 10s of microseconds.

The nastiest problem I ever traced this way was a ethernet card on a bsd unix 
box that would occasionally drop interrupts when the system was under load. 
Being able to have two machines looking on the wires while the problem machine 
was acting as a router was the only way I found it. The events rarely lasted 
more than a couple milliseconds, so 10+ms timing just wouldn't have been good 
enough. Someone might have been able to guess the problem other ways, but with 
this I had tools to see the problem.

So usually you don't need it, and if you don't have it you won't even believe 
you need it, but I have used it and always want it available. Not many (maybe 
none) customers are going to care below the 10s on milliseconds, but as a 
network engineers it could save you.

(end or rant)

jerry




Current thread: