Re: New Descriptor.]

["Thomas 'Dent' Mirlacher" <dent@cosy.sbg.ac.at>]

On Tue, 23 Apr 2002, Patrick Cipiere wrote:

> 
> "Kearney" <clausen@cosy.sbg.ac.at> wrote:
> 
> > I think this is absolute overkill - to address 2^28 multicast groups or
> > individual stations in ONE PID stream is unrealistic. And remember: every
> > bit counts.
> 
> How many addresses do you think is realistic?
> I don't understand why we should not allow the full IP adresses scope
> on one PID.

another related question would be: 

how does the correct scheme for segmenting the IP-address space
into PIDs look like.

and there are a couple of possible solutions (also remember there
	are receivers out there which are not capable of filtering
	2^13 PIDs)

1) put all the traffic in one PID
	+ simple implementation on the receiver side (just filter a
		single PID)
	- load on the receiver

2) put unicast inside one PID
   group multicast in several other PIDs
	+ (see above - at least for P2P connections)
	- for the unicast case, there's no easy way to
		do fast filtering (ar HW/FW level, without
		looking at the IP address - so the work
		needs to be done in the IP stack)

3) put every connection into a single PID (like PVCs)	
	- PID space exhaustion: possible problems with OBPs (like skyplex)
	- most existing receivers cannot HW-filter more than n-PIDs

all the above points share the need for an IP->PID table.

	tm
-- 
in some way i do, and in some way i don't.