🚀 go-pugleaf

I am enjoying my exploration of time.  I have some questions that I'd like  
to ask.

Here at McCaw I'm proposing a three tier time network.  There are 39 hosts  
synchronizing time across the WAN.  Four of them are connected to either  
receivers or Internet secondaries.  Those hosts are peered with one  
another and are our stratum 1 servers.

The remaining 35 hosts are stratum 2.  Each host is peered with one  
stratum one, a client of the remaining three stratum ones, and peered with  
two other stratum 2 hosts.

Each server is located at a well know router site where many user hosts  
connect.  Each user host will be a client of a maximum of four servers.   
I'm currently looking at load and traffic to determine if this is a  
reasonable solution.  Because the router sites are geographically  
dispersed and there are many cooperating servers, it seems like a  
reasonable configuration.

I'm considering using Sun somethings as the servers.  I have calculated a  
worst case poll rate of four queries per second to these hosts.  One  
question I have for the net is load.  The configuration doesn't feel like  
a significant network or CPU load.  What is your opinion?  (For my  
calculations I assumed sixty four user hosts at each router site.)  Would  
it be better to use computers from a single vendor or multiple vendors as  
the servers?  

McCaw has made a significant investment in NeXT computers.  Those  
computers will be user hosts.  Many other computers will be user hosts too  
such as PeeCees, Macintoshes, Suns, HPs, etc.  From the Unix perspective I  
have a clue.  But another question I have is what packages are available  
for the PCs and Macs such that they can time sync too?  

What commercial packages are available to implement NTP?  There is an  
issue regarding support.

What are the PD NTP packages?  I'm aware of xntp3 and ntp-test.  Are there  
others?

In one of my previous postings I had a question about loops.  What is a  
time loop and why is it bad?  Is a time loop where two hosts are peered  
with each other or where a group of hosts are peered with each other? 

Regarding NeXT:
NeXT ntp seems to ignore /etc/ntp.conf and suck its configuration from  
NetInfo.  The ntp configuration files I've seen contain three server  
specification types:  peer, server, and passive.  NeXT has implemented  
_server_, _host_, and something else which I haven't looked into.  There  
must be some magic code in their modified version of ntp that looks at  
those entries and host name and comes to some magical configuration  
conclusion.  I have only one problem with this:  no documentation.  The  
documentation that does exist says something to the effect:  "Use  
HostManager.  It just works."  While I'm sure this is reasonable in a  
small network, bigger networks have bigger issues and there must be some  
definition management.  My question ultimately comes to: does anyone have  
a reasonable explanation of NeXT's implementation?  The issue is  
controlling load distribution.  If a computer isn't listed in one of the  
NetInfo entries then, does it poll all of the hosts listed?  (BTW, we're  
3.0.)

Finally, are there any hacked versions of ntp that deal with Kerberos on  
the peer level?  Xntp3 implements a form of authentication but this form  
is less manageable across our WAN than something that uses Kerberos.   
Kerberos and ntp seems like an ugly mix.

--
Dennis P. Glatting / Sr. Technical Manager / Trirex Systems Inc.
315 Post Road West / Westport, Connecticut 06880 / (203)221-4600
dennis_glatting@trirex.com (NeXTmail Ok)
Member League for Programming Freedom