From: harrison@beta.lanl.gov (David A. Harrison)
Subject: Re: SCI: Timestamps...
Date: Mon, 10 Aug 1992 18:06:15 GMT
Message-ID: <1992Aug10.180615.18867@newshost.lanl.gov>
Organization: Los Alamos National Laboratory


In article <1992Aug10.040724.20404@u.washington.edu>
salnick@dejavu.spk.wa.us (There is no substitute for displacement...)
writes:

>In article <1992Aug3.202012.26129@newshost.lanl.gov>
>harrison@beta.lanl.gov (David A. Harrison) writes:
>
>>Back to timestamping packets...Does anyone know of a way to
>>synchronize the clocks of many computers to allow useful timestamping.
>>I don't think that its possible (at least not with any reasonable
>>method).  I don't think that timestamping packets is necessary. As
>
>At work, we have a network with 60 VAXen on it (and hundreds of other
>computers...)  - all the VAX system clocks are kept to within 10
>milliseconds of each other.  Proof by existance that it is possible.
>Doing this uses no detectable bandwidth or CPU.

I may have been unclear in previous messages.  I agree that
timestamping is possible on the scale of a single LAN.  I am impressed
that 10 millisecond accuracy can be maintained on 60 computers with
little affect on bandwidth; however, this does not answer the
possibility of synchronizing on a WAN.  If you want to synchronize on
a WAN then I think that some other techniques are necessary.  I would
think that causality is only a problem once you get to the WAN scale
unless the virtual reality is running some very time critical
simulations.  For human interactions causality problems would be
negligible on a LAN making timestamping unnecessary for most
applications--although I won't say unnecessary for all applications.

David Harrison...