RE: CRC

• Date: Tue, 13 Feb 1996 14:07:29 PST
• From:
  Roger Crew <rfc@microsoft.com>
• Cc: "moo-cows@parc.xerox.com" <moo-cows@parc.xerox.com>
• Encoding: 30 TEXT

    Roger> (* and possibly also which MOO the code was first written
    Roger> on, if you want to be able to have alterations flow in both
    Roger> directions, but you have to be careful about the situation
    Roger> where the clocks are not entirely in sync)

> I would suggest that NTP (Network Time Protocol) be required to use
> this mechanism, and that everyone sync on an agreed upon master,
> perhaps PARC.xerox.com, or the national beureau of standards, or some
> other known source.

It's not so easy.  Even with NTP, given the vagaries of hardware clocks,
system load, relativity, and other factors, you're NEVER going to get
perfect synchronization; consider it a fact of life that the clocks will
always be a bit off.  How much depends on how well the machines in your
vicinity are administered, how heavily they're loaded, etc... this could be
a few milliseconds, a few seconds, or even a few minutes (e.g., if you're
like me and live on a subnet where they've never even heard of NTP).  

Thus some thought needs to be given to the situation where a version of the
verb arrives with a timestamp from a different site and a few seconds off
(particularly, a few seconds *earlier*) than the one you already have. 
Saving the losing version in a log and posting a warning may suffice.

BTW - PARC.xerox.com would be a particularly *bad* choice as time master. 
The ideal time master is a very *lightly* loaded system with a good hardware
clock and a serial port listening to WWV.  And you want to have lots of
these scattered all over the place so that no one timeserver need talk to
too many clients (... every millisecond you spend servicing a client is one
you don't have for watching the WWV port and making sure your clock is still
in sync).

• Prev: RE: CRC
• Next: Re: CRC
• Index: MOO-cows
• Thread: MOO-cows