On Tue, 11 Oct 2016 08:43:46 -0700, Charles Mills <[email protected]> wrote:
>Okay, I read the paper. Good paper. Thanks. (I don't feel too bad -- the
>paper only came out a couple of months ago.)
>
It's hardly automatic; according to the reference:
Lastly, through the Hardware Management Console, manually schedule STP to
insert the new leap seconds when they are announced and due to occur.
"schedule" implies a comfortable lead time.
>The paper is short on low level details. Does anyone know: in a "steering"
>situation, does z/OS basically "change CVTLSO a little, spin a little, go
>back to business as usual for a while -- repeat as necessary for seven
>hours"?
>
Any clock will drift with respect to a standard. STP compares the ETOD clock
with an external standard and loads a steering register that adjusts the ETOD
rate in a manner that the CPU never perceives a discontinuity. At a leap second
STP observes that ETOD is a second fast and adjusts the ETOD rate to the
minimum. It takes about a third of a day to fall into sync.
In the non-steering case, all (user?) processes are non-dispatchable for
one second. This could pose a problem for certain (obsolete?) peripheral
devices. Ed Gould commented on this several months ago.
Both Amazon and Google have conventions to spread the leap second
over several hours centered on midnight. Unsurprisingly, they chose
different durations for the spread.
>-----Original Message-----
>From: Mark Jacobs - Listserv
>Sent: Tuesday, October 11, 2016 7:21 AM
>
>STP will recognize that a leap second has occurred, and slowly steer zOS
>time to the actual time. Here's a good writeup on the process.
>
>https://www-03.ibm.com/support/techdocs/atsmastr.nsf/WebIndex/WP102081
-- gil
----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN
