underh20.scubadiving@gmail.com wrote:
> We have few Solaris 8 and 10 servers. Does anyone know how to setup a
> central NTP server for all our other servers to synch. the clock
> with ? We don't have any budget for a dedicated NTP server. What's
> the best practice in setting up the NTP server ?
>
> Thanks,
>
> Bill
Select four to seven internet NTP servers that are near you in net
space. "Near you" basically means a low number for "delay"; anything
less that 20 milliseconds is generally good enough. A quick and easy
test is to "ping" each candidate and see what the round trip delay is.
This is important because the maximum error in transmitting time from
server to client with NTP is one half the round trip delay between the
two systems.
You will find lists of stratum one and stratum two servers at ntp.org.
You should normally use stratum two servers unless you are serving more
than 100 clients.
Your configuration file is /etc/inet/ntp.conf. There are a couple of
samaples in there.
The version of ntpd that ships with Solaris is eight or ten years old.
There has been a lot of development since. You might want to download
the reference implementation from ntp.org and build the executable from
source.
You will need to set your clock to a good approximation of the correct
time before starting ntpd. If the clock is off by more than about 17
minutes, ntpd will take a panic exit!!
The closer you can set it, the faster ntp will get it synchronized.
Using the version shipped with Solaris, you need to use ntpdate to set
the clock. The reference implementation has a "-g" option which causes
ntpd to set the clock on a one time basis.
You want to start ntpd BEFORE you start anything that depends on
monotonically increasing time. Some software can get more than a little
upset if the time steps, especially if it steps backwards. Once past
the initial startup, ntpd will normally only correct the clock by
running it a little faster or a little slower.
If you use the reference implementation, one of the improvements is the
"iburst" keyword for the server statements in ntp.conf:
server glassclock.badschuschein.org iburst
server . . . . iburst
server . . . . iburst
..
..
..
Iburst tells ntpd to send the first eight request packets at two second
intervals. This gets ntpd enough information to start disciplining the
clock in the first sixteen seconds. Once the initial burst is complete,
ntpd will poll the servers at intervals between 64 and 1024 second.
Ntpd selects the proper polling interval for the conditions then
obtaining! Just let it do it's thing and don't worry about it; it's
much smarter about such things than any human. My over simplified
expanation of the math I don't really understand is that the shorter
poll intervals are used to correct large errors quickly while the longer
intervals are used to correct small errors very accurately.
There is a lot of useful documentation and advice available at ntp.org.
Go and look at it!
You can use "ntpq -p" to monitor ntpd's performance. Ntpd may need as
long as thirty minutes to beat your clock into submission so don't take
any ntpq -p output very seriously until ntpd has been running for at
least thirty minutes.
One other thing. Sun still calls the daemon "xntpd". The "x" stood for
"experimental" and the developers dropped it many years ago.
