Andr? van Toly <[EMAIL PROTECTED]> wrote:
> At 13:16 +0200 20-04-2004, Michiel Meeuwissen wrote:
> >There are more problems with multicast:
> >
> >- IIRC a multicast is not garanteed to be received
> 
> That's true, multicast relies on UDP (like streams) in stead of TCP.
> 
> >- multicast can only be done within the same subnet, while this is not true
> >  for the database connection. So, if you run 2 mmbase's on the same 
> >  database,
> >  but in a different network, multicast cannot be used to synchronize the
> >  caches.
> 
> OK, but creating something with for instance RMI 
> (TCP based) would need more resources...

Since multicast only happens at editing, peformance should not easily be a
problem. I think that even if you set it up with one Thread it should easily
be able to handle quite a lot of edits. And furthermore a delay of a few
seconds does not matter, do it can simply queue if a 'burst' of invalidation
requests comes in, and handle them one-by-one.


> >So, perhaps there needs to be a small project who offers an alternative for
> >multicast. Perhaps using RMI or so. It should then of course also consider
> >this Thread issue.
> If i remember reading correctly at irc this 
> afternoon is that more MMBase modules suffer from 
> this issue.

Perhaps, but I would not know where excactly it could be problematic.

I do btw think also that it is a bit silly that Threads are so expensive.
The way to avoid the threads now, is writing your own code doing more or
less the same, only in a specialized way (e.g. FileWatcher which now has
only one static Thread, which basicly is a loop dividing the work to the
FileWatchers instances, which sounds to me as a simply Thread simulation)

 Michiel


-- 
Michiel Meeuwissen       |
Mediapark C101 Hilversum | 
+31 (0)35 6772979        |  I hate computers
nl_NL eo_XX en_US        |
mihxil'                  |
 [] ()                   |

Reply via email to