Re: [GnomeMeeting-devel-list] ILS, big problem

[Damien Sandras <dsandras seconix com>]

Le lun 13/10/2003 à 18:50, Kilian Krause a écrit :
> Hi Damien,
>  
> > > if you scan through the ils list on a difference basis (i.e. only
> > > changed and new users since last scan), you could just flag those users
> > > with an ldap field that their ip cannot be called. that way you'd have
> > > say gray, red and green users in the ils browser. 
> > 
> > It wouldn't be compatible with Netmeeting and old GM versions.
> 
> well, we could just edit the comment field in that case and insert a
> high-bit char as #1 or as last. nobody will be annoyed, but GM does know
> it's a broken config. 
> 

Why not simply removing those users? Why are they registered to ILS if
they can't be called?

Anyway, what you propose is to add to the server a thread that
separately scan the list of new users every x minutes to flag or delete
unreachable users.

The problem with this approach is that :
- scanning takes a very long time, even if you scan only new users
- the server is in perl (I don't know if threads exist in PERL)
- it could add more instability to the server

In that case, the easiest still seem to do the following :
- when the user registers, check if he is really new or not
- if he is really new, check if the ip is reachable

That is easier to do, and probably less CPU intensive. Yes, it could
lead to a DOS, but registering 1000 spoofed IP's to the server would
also DOS it without that, so is it worth to try creating threads for
what we want to achieve?

> > > the gray ones are callable but busy.
> > > the red ones are not callable at all.
> > 
> > Why display them then?
> 
> to "explain" to newbies why those can't be called. it'll decrease the
> "well, i can call *NOBODY* on ILS" messages i hope.
> 
> > > and the green ones are callable and not busy.
> > > 
> > > if then gm would scan for this field set every once in a while (when
> > > registering/unregistering on the server e.g.) a popup could be displayed
> > > that a regular check has proven the setup to be non-working. 
> > > 
> > > at least for a non push-processing system (i.e. server can tell client
> > > directly) that seems to work fine at reasonable load to me.
> > > 
> > > Objections?
> > 
> > What is the advantage over solution 3?
> 
> that it's no active server echo. so no client or fake client can DOS the
> server with requests that way.
-- 
 _	Damien Sandras
(o-	GnomeMeeting: http://www.gnomemeeting.org/
//\	FOSDEM 2003:  http://www.fosdem.org
v_/_	
	H.323 phone:  callto://ils.seconix.com/dsandras seconix com