[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Time Management and Nullmessages
From: |
Pierre Siron |
Subject: |
Re: Time Management and Nullmessages |
Date: |
Mon, 04 Feb 2008 10:24:22 +0100 |
User-agent: |
Thunderbird 2.0.0.5 (X11/20070719) |
Christian Stenzel a écrit :
> Hello all,
>
> for the preparation of my talk in Magdeburg I've read some
> publications dealing with CERTI. My main resource is
> ftp://ftp.cert.fr/pub/siron/.
>
> As I read in ftp://ftp.cert.fr/pub/siron/02F-SIW-018.pdf
> the time synchronization relies on the original Nullmessage-algorithm
> of Chandy and Misra, also called CMB-algorithm.
>
> There a nullmessage is a promise not to generate
> any message with a lower timestamp then tnow+lookahead. Nullmessages
> are generated after each increment of the simulation clock (I suppose
> after each TAR resp. NER, if no "normal" message is available).
>
hello Christian,
yes, and ...
this can be worse (we can send a Nullmessage after the reception of another
and go into the "time creep" situation if the lookahead value is too small)
> As I think to know this algoritm do not allow something like a
> zero-lookahead. When using zero-lookahead, normally it should lead to
> an infinite synchronization loop. Also depends the performance of the
> distributed simulation highly on the value of the lookahead and
> with it on knowledge depending on the concrete simulation model.
> Small lookaheads generate much more nullmessages and network traffic.
>
> I've done no experiments concerning zero-lookahead and CERTI yet, so
> perhaps anyone of you had. This matter directly influences the TARA
> and the NERA service, I suppose.
>
we plan to implement TARA and NERA services (#6898 task)
maybe with a shortcut in the RTIG process
> As I know protocols like Deadlock Detection and Recovery (Chandy, Misra)
> or synchronous variants can handle zero lookahead.
>
yes
this is the second generation of time management algorithms
> So the question is if I could use zero-lookahead with CERTI.
>
>
>
not yet
can we implement easily a deadlock detector in the RTIG ?
we have to study this question
bien cordialement
Pierre