On Thu, 24 May 2007 08:36:47 -0500, Miller, Pat <[EMAIL PROTECTED]> wrote:

> ..., we are going to set up a script to capture this time and issue
>console SET CLOCK commands.  May be off by a few seconds, but good
>enough for our humble purposes at this time.
>
This will get the local time correct, probably better than "a few seconds",
perhaps even a fraction of a second.  But it leaves GMT, the Gold Standard
for critical timestamps drifting aimlessly.

I've wondered about the possibility of a process which would make small
adjustments repeatedly to CVTLSO.  This should adjust both local and GMT,
leaving out only jobs that issue STCK directly but ignore CVTLSO.

>In the future, we will consider STP and sync to the same source as the 
>network, which might eventually evolve to option 1, below (those of you who 
>have fought in the Architecture Wars know you have to choose your battles 
>carefully).
>
The first UNIX system on my desktop was a Sparc (I forget the model) running
SunOS 4.3.  It had a privileged command (and I had the privlege), "date -a"
which was documented to tweak the clock frequency by a small amount until
the specified time adjustment was complete, then restore it to standard rate.
The man page said, "a fraction of one percent".  My experiment showed it
was more like 25%.  Whatever.  It worked, and I don't believe the clock was
ever observed to run backward.

The description of the hardware for STP mentions similar facilities.  There's
a clock offset register, which is added to the value returned by STCK, and
a rate offset register which is periodically added to the clock offset register.
The maximum slew is comparable to that of the Sysplex Timer, a few seconds
a day, not enough for some purposes.  While a 25% overdrive or underdrive of
the TOD clock would be considered harmful in many cases, surely it would be a
boon to the Eastern Hemisphere site which is desperately seeking a way to
convert from TOD=Local to TOD=GMT without shutting down for several hours.

-- gil

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO
Search the archives at http://bama.ua.edu/archives/ibm-main.html

Reply via email to