On Tue, 11 Oct 2016 10:39:20 -0700, Charles Mills wrote:

>Ah! I am starting to understand. Leap Second steering is accomplished with
>the PTFF instruction and is independent of CVTLSO. PTFF appears to slow down
>the physical clock.
>
>So ... steering and CVTLSO are essentially alternatives, right? CVTLSO
>should not include any leap seconds that were previously (or were about to
>be) "steered" -- is that right? And if a shop is using STP it is probably
>not modifying CVTLSO: CVTLSO is probably either zero, or at least stable. Am
>I getting this right?
> 
CVTLSO should be the difference between ETOD and UTC regardless of how
prior leap second adjustments were accomplished.

If a shop chooses to keep CVTLSO zero, STP will steer ETOD to match UTC
within a few hours after a leap second.

If a shop chooses to use CVTLSO non-zero, STP (perhaps by non-disclosed
techniques) will be signalled by the HMC at a leap second to make user
tasks non-dispatchable for a second during which the CPU will add one
second to CVTLSO.

In principle, a shop could elect to run ETOD at TAI, always ten seconds
ahead of IBM's recommendation and maintain CVTLSO at a corresponding
ten second higher value.  But why?

It's not clear why the convention is to add CVTLDTO to ETOD but subtract
CVTLSO.  Perhaps a designer had a phobia of negative numbers.

I believe either Peter Relson or John Eells has explained all this with more
detail and probabaly more accurately than I in some previous June or
December.

-- gil

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN

Reply via email to