Re: Time Synchronization

New Message Reply About this list Date view Thread view Subject view Author view Attachment view

From: Andy Hisgen (hisgen@caribe-85.eng)
Date: 03/11/97-11:34:57 AM Z


Date: Tue, 11 Mar 1997 11:34:57 -0600 (Central Standard Time)
From: Andy Hisgen <hisgen@caribe-85.eng>
Subject: Re: Time Synchronization
Message-ID: <Roam4.0.5.858101697.28682.hisgen@caribe-85.eng>

I'm rusty, but I'm not convinced that time synchronization is
absolutely necessary even for embargo resolution.  It should
be sufficient (and necessary) to know that the difference
in the clock rates between client and server, i.e., the
difference in the rate at which their clocks advance, differs
by a known upper bound.  There should be no need for them
to agree on what the actual absolute time is according to
God (or the British Navy).

In practice, that bound can be large, e.g., a factor of two.
And it should be possible to design the protocol so that
a client whose rate is wildly out of wack does not affect
the correctness of the algorithm -- that client may not
get any useful work done though.

Caution: Even knowing that the clock rates differ by a known upper
bound is problematic with very heterogenous clients.  E.g., consider
a notebook computer that goes to sleep.


>Following on the discussion at Connecathon, upon checking the Spritely NFS
>paper, I find that it does indeed require time synchronization between
>client
>and server, in particular for embargo resolution.
>
>It may be inevitable the if NFS V4 is going to include some enhanced cache
>consistency abilities, it will require some form of time synchronization. 
>My
>initial feeling is that this requirement may be incompatible with the desire
>for
>wide spread deployment of WebNFS clients, particularly on home PCs.


New Message Reply About this list Date view Thread view Subject view Author view Attachment view

This archive was generated by hypermail 2.1.2 : 03/04/05-01:45:30 AM Z CST