Re: scheduler!!

[Polly Huang <huang@tik.ee.ethz.ch>]

Martin,

'going backwards' doesn't usually interupt a simulation run.  The direct
cause of failuer is rather a 'floating point problem' (though may be
a result of the numeric stability problem in Calendar scheduler).

ns: _o813 recv useful 1 13 10: floating-point value too large to represent
    while executing
"format "%7.4f [[$self set node_] id] $op-$type $args" [$ns_ now]"
[...]

How about changing %7.4f to %20.6g in tcl/mcast/mftp_rcv.tcl's evTrace
instproc?  This might allow the simulation to continue but will
probably dump out funny traces.  To eliminate the 'going backwards'
warning all at once, I can only think of two other possible solutions
(given that you've tried Heap scheduler):

1. Change to List scheduler (the most stable one I suppose) with this:
$ns_ use-scheduler List
2. Download and recompile the latest scheduler code as Yuri suggested in
the message:
http://www-mash.cs.berkeley.edu/dist/archive/ns-users/9909/0102.html

If the 'going backward' problem persists and depending on how much
time it is going backwards (look at the funny traces), you might need to

1. dig into the mftp code and find out where events are schduled to 
be executed in earlier time. (sorry, you'll be on your own if this
is the case.  Although mftp is integrated in the distribution, it's not
currently supported by the ns developement team) 
2. or report a bug in List scheduler code (which I think is very
unlikely...)

hope this helps,
-Polly

On Mon, 7 Feb 2000, Martin Koyabe wrote:

> Hi Polly,
> 
> Thanks for the correction that I received !! I saw the mistake I made in my 
> earlier code !!
> 
> I'm currently getting the scheduler problem (see below) when I increase the 
> number of nodes !! Ofcourse, this could be due to the 'uneven' arrival of both 
> repair packets and NACKs to my MFTP/Snd agent. When I look at the 'mailing-list 
> threads' on scheduler problems:
> 
> http://www-mash.cs.berkeley.edu/dist/archive/ns-users/9909/0102.html
> 
> The solution suggested seems not to work even after using Heap instead of 
> Calender (default) scheduler.
> 
> Any leads/suggestions ?
> 
> Tx
> 
> -- Martin
> 
> e-mail: koyabe@erg.abdn.ac.uk
> 
> -------Error-----
> Creating links for WAN ....
> Creating links for LAN with (1..10) nodes...
> Creating links for LAN with (10..20) nodes...
> Creating links for LAN with (20..30) nodes...
> Inserting link errors (1..10) nodes...
> Inserting link errors (10..20) nodes...
> Inserting link errors (20..30) nodes...
> Transmitting 2000000 byte file with group-size 8
> simulation is running...
> ns: scheduler going backwards in time from 1.288963 to 0.000000.
> ns: _o813 recv useful 1 13 10: floating-point value too large to represent
>     while executing
> "format "%7.4f [[$self set node_] id] $op-$type $args" [$ns_ now]"
>     (procedure "_o813" line 4)
>     (Agent/MFTP/Rcv evTrace line 4)
>     invoked from within
> "_o813 evTrace recv useful 1 13 10"
>     ("eval" body line 1)
>     invoked from within
> "eval $self evTrace $proc $type $args"
>     (procedure "_o813" line 2)
>     (Agent/MFTP/Rcv recv line 2)
>     invoked from within
> "_o813 recv useful 1 13 10"
> 
> ---------------------------------------------------------------------
>