[comp.protocols.time.ntp] How widely deployed is NTP?

ellozy@FARBER.HARVARD.EDU (Mohamed Ellozy) (03/08/91)

Dave,

The Internet is probably too big today for another "glorious experiment",
but do you have any "feel" for how widely NTP is deployed?

Do any vendors, other than DEC and TGV, ship it with their software?

Mohamed

Mills@udel.edu (03/08/91)

Mohamed,

I'm guessing at 5000 clients now, but HP and Sun both say they
have "thousands" of machines on their internal net and believe
most run NTP at least on reboot. Sun has announced plans to
ship it in the next release and I hear TWG is doing that for
Multinet.

Dave

ken@HPSDLZ.SDD.HP.COM (Ken Stone) (03/08/91)

> I'm guessing at 5000 clients now, but HP and Sun both say they
> have "thousands" of machines on their internal net and believe
> most run NTP at least on reboot. Sun has announced plans to
> ship it in the next release and I hear TWG is doing that for
> Multinet.

Dave ...

I've been running a sort of "glorious experiment" within net 15 (HP)
to try and determine how widespread the use of NTP really is.  The
results have been a little surprising.  While I expected to see on
the order of 2500 to 3000 hosts using ntp,  what I found was a varying 
number that seems to run from 500 to 750 on the outside :-(.  Its kind of
a top heavy tree with as many as 30 strata 2's .... more data here
as I determine it.  The reason for this project is that we are working 
on a more optimal internal structure.

Tools used include a mongo perl script and some mod'd binaries of ntp,
xntpdc, and ntpq.

  -- Ken

bdale@COL.HP.COM (Bdale Garbee) (03/08/91)

> I've been running a sort of "glorious experiment" within net 15 (HP)
> to try and determine how widespread the use of NTP really is.  The
> results have been a little surprising.  While I expected to see on
> the order of 2500 to 3000 hosts using ntp,  what I found was a varying 
> number that seems to run from 500 to 750 on the outside :-(. 

I may be presumptuous to assume that I'm a typical HP internal site admin, but
I have a multi-hundred machine site that is adopting NTP, and want to offer
a thought or two on the subject.

We've put up a stratum 2 and several stratum 3's on our site, and have turned 
on a few diskless servers at stratum 4... we ran into some initial 
configuration and learning problems which have since been resolved, and expect
to turn on all remaining standalones and diskless servers at stratum 4 "real
soon now".

But even then, the numbers will seem low since we run a *lot* of diskless 
nodes, and HP's diskless code already does good time synchronization between 
diskless nodes and their servers.  So we'll only have the servers running 
NTP...  therefore the number of machines "synchronized by NTP" will really
be an order of magnitude larger than the number of machines "running NTP"
given our average diskless cluster size.

Not sure anyone cares, but it always helps to know what you're counting...

Bdale, N3EUA

tengi@princeton.edu (Christopher Tengi) (03/08/91)

In article <9103071709.aa15122@huey.udel.edu>, Mills@udel.edu writes:
|> Mohamed,
|> 
|> Sun has announced plans to
|> ship it in the next release and I hear TWG is doing that for
|> Multinet.
|> 

Dave,
    FYI, NeXT's 2.0 release also includes an ntpd.  It is configured
using their netinfo system.  I don't handle the NeXTs myself, so I
can't give any details on how they get configured, but I believe that
the daemon is derived from the ntp distribution on louie.udel.edu
(whenever NeXT took their snapshot).

-- 

==========----------==========---------+---------==========----------==========

	UUCP:	  ...princeton!tengi		VOICEnet: 609-258-6799
	INTERNET: tengi@princeton.edu		FAX:      609-258-3943
	BITNET:	  TENGI@PUCC

anders@ifi.uio.no (Anders Ellefsrud) (03/09/91)

In article <9103072309.AA19633@hpsdlz.sdd.hp.com>, ken@HPSDLZ.SDD.HP.COM (Ken Stone) writes:

> I've been running a sort of "glorious experiment" within net 15 (HP)
> to try and determine how widespread the use of NTP really is.

This inspired me for a little experiment (Norway is still small enough).

I dug out all A-records from the no domain (Norway). All this hosts, a
total of 5912 addresses, was sent an ntp packet. This included a bunch
of routers, PC's, mac's, terminal-servers and other random equipment
which I would not expect to respond to ntp packets anyway. I ran the
test just once, so hosts beeing down or othervise unable to respond at
the moment they where probed could skew the numbers slightly. Things
with more than one ip-address was tested mulitiple times which also
tend to introduce inaccuracies.

Here are the numbers.

Hosts running ntp:      466
Hosts not running ntp: 5446
Total number of hosts: 5912

This indicates that about 8% of all registered ``things'' with an
ip-address in norway was running ntp and responding to ntp requests.

--
anders@ifi.uio.no

syd@DSI.COM (Syd Weinstein) (03/09/91)

When talking about all the hosts running ntp, what consideration
is made for those that don't need millisecond accuracy and just
run ntpdate every couple of hours?

All of our systems run ntpdate very hour (if their clocks drift some)
or every three hours if they don't drift much.  It keeps them all
withing a second quite nicely.  That seems close enough for what
we do, and of course we use ntpdate at boot time to set the clock.
-- 
=====================================================================
Sydney S. Weinstein, CDP, CCP                   Elm Coordinator
Datacomp Systems, Inc.                          Voice: (215) 947-9900
syd@DSI.COM or dsinc!syd                        FAX:   (215) 938-0235

Mills@udel.edu (03/09/91)

Anders,

Well, if that fraction holds throughout the Internet, there should
be some 24,000 chimer and chimee NTP hosts. I wouldn't be surprised,
since the level of the assault on the fuzzball primary servers has
easily doubled over six months ago.

Dave

louie@SAYSHELL.UMD.EDU ("Louis A. Mamakos") (03/09/91)

The NeXT ntpd implementation is derived from the ntp 3.4 version that we
wrote here at the University of Maryland.  I also have some changes that 
they made to interface with the Netinfo system they use.  It seems confined
to the part of ntpd that reads and parses the configuration file.  Otherwise,
it seems to be a rather standard version of ntpd.

My personal thanks to NeXT for tracking down and killing horrible kernel bugs
that made the NeXT 0.8 version of the software behave almost as bad as SunOS
does.  It seems to keep rather good time.

louie

ken@HPSDLZ.SDD.HP.COM (Ken Stone) (03/09/91)

> All of our systems run ntpdate very hour (if their clocks drift some)
> or every three hours if they don't drift much.  It keeps them all
> withing a second quite nicely.  That seems close enough for what
> we do, and of course we use ntpdate at boot time to set the clock.

I ran some tests here and found that an xntpd running as a broadcast 
client took less umph than running ntpdate or rtime or whatever out 
of cron.  We found that we had large cad program users that could tell 
us when ntpdate ran out of cron consistantly even with us varying the 
times whereas a xntpd presented a much lighter and consistant load that 
no one noticed.  This is on medium sized workstations.

Just a thought ...

  -- Ken