On 5/20/2014 1:09 PM, Mike Skvarenina wrote:
Since we're using an HMC and the System i
connects to it more or less we are going to show the HMC console/NTP setup
and the System i log files showing how the time is adjusted automatically
which should be acceptable.
I haven't got an HMC. Does the HMC reach into the Power 7 and set the
clock? Or does it require SNTP to be running on the IBM i side?
I need to meet a government requirement that says
my System i must use the NTP protocol and set its time.
What is the requirement? The government has a document for everything.
It matters because:
I'm well aware of IBM's built in support for
SNTP but the requirements state that SNTP ("simple"
network time protocol) is not sufficient.
One of the reasons given is that SNTP allows you to
define multiple time servers on the Internet but it
uses a round robin approach such that if one server
is down it goes to the next. With NTP,
apparently multiple servers are considered by the
protocol and it determines which is the most accurate.
NTP does average several time servers in order to get closer to UTC, but
the really big difference between the two is how they treat time
changes. Imagine you have a network outage for a few seconds and NTP
finally reaches a time server and gets a new packet. It discovers that
the time coming in is different from the system clock and needs to
update that system clock. NTP will speed up the clock (or slow it down)
a bit until the incoming packets match the system clock. In other
words, the clock will gradually slew onto the new time. With SNTP, the
algorithm is simple, and doesn't slew the clock, it makes up the
difference in one jump, so there's a lurch when SNTP needs to make a change.
This is why the actual requirement matters. If the requirement is to
avoid large clock changes, you really need to be running an NTP that can
gradually slew the IBM i clock. If the requirement is to be within 100
microseconds of UTC, I'm not sure you'll be able to meet that with any
time server on a public network no matter which client is running.
Depending on your external network, SNTP may be able to hold the system
clock within a few microseconds of a clock maintained by NTP. It just
won't be smoothly changing.
Finally, something the requirement may not mention: what about the rest
of your data centre? Often, it's more important that all of the clocks
in the machine room are maintained by a single master clock, whether it
be a local stratum 1 server or the USNO via the internet. In a world
where hundreds or thousands of transactions per hour hit multiple
machines and databases, it pays to have all those databases synchronised
with each other time-wise. This may not be an issue in your particular
case, but it might be an issue for someone reading the archives.