NetNews Usenet Archive 1992 #19/NN_1992_19.iso/spool/sci/logic/1342

Text File | 1992-09-03 | 3.9 KB | 100 lines

Newsgroups: sci.logic
Path: sparky!uunet!destroyer!gatech!hubcap!fpst
From: [e-mail] (Steve Stevenson)
Subject: Re: Synchonizing watches
Message-ID: <[e-mail]>
Organization: Clemson University
References: <[e-mail]>
Distribution: usa
Date: Thu, 3 Sep 1992 17:26:32 GMT
Lines: 88

[e-mail] (Doug Stevens) writes:

>------
>A and B are two people at two ends of a very slow telephone connection.
>They are both completely cut off from the world except for their connection
>with each other.  In particular, they have no idea what time it is in the
>real world.

>There is a delay AB in the transmission of what is being said by
>A to B, and a delay BA in the transmission of what is being said
>by B to A. The delay AB does not equal the delay BA. The delays, for the
>purpose of this problem, are on the order of minutes in each direction.

>Both A and B have watches. Before they begin to talk, the watches are not
>set to the correct time, nor are they set to the same time.

>The problem is to devise a procedure such that A and B can, via their
>telephone connection only, accurately synchronize their watches (to
>within a few seconds). The watches do not have to reflect time in the
>real world, but must match each other. Both A and B will know the procedure
>before the call begins, and each will know whether he is A or B.

I posted this problem to my news group, comp.parallel. Here are two
solutions I've received. The first gives a solution and the second
gives a Thinking Machines Corporation publication which deals with the
problem

=========================== MODERATOR ==============================
Steve Stevenson                            [e-mail]
Administrative address                     [e-mail]
Department of Computer Science,            comp.parallel
Clemson University, Clemson, SC 29634-1906 [phone].mabell
====================================================================

Date: Thu, 3 Sep 92 11:30:41 -0400
From: Clark VERBRUGGE <[e-mail]>
Message-Id: <[e-mail]>
----------------------------------------------------------------------------
- A Solution -
----------------------------------------------------------------------------
Let the local time of A be tA, and the local time of B be tB.
Further, let tAB be the lag between messages going from A to B,
and let tBA be the symmetric value. Since both time pieces
are presumable ticking at the same rate, let tB = tA + C for some
constant C. If we can determine C, then they can synchronize their
watches.

1) A asks B the time. As s/he does, s/he notes how long it
   takes to hear the response from B. If B answers promptly,
   A will now know 3 things:
      (tA)                 (local time)
      (tAB + tBA)          (total delay)
      (tA + C + tAB)       (time told to A by B)

2) A subtracts the last value from the first two:
      (tA) + (tAB + tBA) - (tA + C + tAB) = (tBA - C)

3) B now asks A the time. A hears the request at time
   tA', notes this time and responds. B now knows:
      (tA' + tBA)          (time told to B by A)

4) B tells A the value of (tA' + tBA). A can now calculate:
      (tA') + (tBA - C) - (tA' + tBA) = -C

5) A adds -C to his/her watch. Now both watches are in synch.

Of course this algorithm can only be as accurate as the
reaction time of the participants. It also assumes
that tAB and tBA are constant.

Clark Verbrugge
[e-mail]

====================================================================
From: Barry Margolin <[e-mail]>

Take a look at the algorithms in the Network Time Protocol, RFC 1305.  Or
if the specific algorithms aren't a good match, the references (there are
about 60 of them) at the end of the document should be useful.

--
Barry Margolin
System Manager, Thinking Machines Corp.

[e-mail]          {uunet,harvard}!think!barmar