Conference noted::atm

    To be precise
    
    .0� There is a camera connected to
    
        a NEMESYS box connected to
    
    .0�                                one of the ports (6.3) and a screen to

    Well, if you really want to be precise, there's an ATA between the video
    camera and port 6.3 and an ATV between port 6.4 and the screen...
    
    ...but I don't think we're getting very far this way!
    
    Pedro

    If we use an ABR P-P PVC, or a CBR P-MP PVC (100Mbps) the image passes
    fine. However if we use an ABR P-MP PVC the image isn't at all
    acceptable (even if we set the rate of image transmission at 1 Mbps).
    The PVC is set as 0/240 on input (6.3) and 0/240 on output (6.4).
    The P-MP PVCs only have one branch.


ABR P-P and CBR (P-MP or P-P) are done in hardware.

ABR P-MP requires software help.  I don't know what the throughput will be
but it won't be good.  It may be bursty if the CPU takes time out to do
something else like setup an SVC or think about the tolology when a new
switch in connected up.


I think you have to use CBR if you want P-MP.  That has the obvious
complication of not being able to overallocate the bandwidth on any link.
[posted by Notes-News gateway]

    Hal,
    
    Thanks for your answer. But...
    
    >ABR P-P and CBR (P-MP or P-P) are done in hardware.
    
    Good to know.
    
    >ABR P-MP requires software help. I don't know what the throughput will
    >be but it won't be good.  It may be bursty if the CPU takes time out to do
    >something else like setup an SVC or think about the tolology when a new
    >switch in connected up.
    
    Why does it require software help and what kind of help is that? What
    about if the GS/ATM isn't doing anything else (we could disconnect
    everything else)? Will this situation change when we'll support ABR
    with a garanteed bandwidth?
    
    Thanks.
    
    Pedro

In the current generation of linecards there is no hardware mechanism for
replicating cells that are sent across the backplane. Without cell 
replication, the linecard has to generate each cell individually and then
re-transmit them as many times as there are p-mp VCs. Each time the cell
is re-transmitted it must be modified with a new VC. This process is
handled by the micro on board the "source" linecard.

If i recall, and i ask others to corrobarate this, the QLCv2 can inject
cells at the rate of 3-4 Mbit/s via the microprocessor. I believe that this
bandwidth is divided into the number of p-mp VCs supported, so 2 p-mp VCs
would have about 1.5-2 Mbit/s available to each, etc.

Hope this helps,
bill

  Answer .-1 will probably confuse people.
  Gsw/ATM line cards do indeed have the ability to replicate cells in
  hardware as they cross the backplane.  But only for CBR VCs.  Thus
  hardware point-to-multipoint service is provided only for CBR P-Mp VCs
  at present.  A few megabits per second of outbound p-mp service can be
  emulated by the line card processors, but that's too slow for most
  applications.

  A third generation of line cards will increase our set of p-mp services,
  among other new features.
  Any time someone mentions "ABR P-Mp" service, they are really talking
  about "UBR P-Mp" service.  No ATM switch has the ability to merge RM
  cells from multicast branches toward the p-mp source.  Thus the Forum
  ABR service is strictly impossible for p-mp VCs with today's switches.

    Hi again!
    
    In .0 when I wrote "Are we trying something wired or is there a
    problem?", of course it is wired, but I meant weird ;-)
    
    Re: .5 
    
    I don't really understand the difference between a P-P ABR and a
    P-MP ABR with MP=1 (except the hw/sw difference which difference I
    could clearly see...) Should there be a difference between using a QLC
    V1.5 and a QLC V2.0?
    
    Thanks.
    
    Pedro
    

In article <[email protected]>, [email protected] (Hawkeye - Network Support @GEO) writes:

>     Why does it require software help and what kind of help is that? What
>     about if the GS/ATM isn't doing anything else (we could disconnect
>     everything else)? Will this situation change when we'll support ABR
>     with a garanteed bandwidth?


It requires software because the hardware can't handle that case.  The GS/ATM
is a crossbar (not a time-shared bus).  The tricky part of building a 
crossbar is figuring out how to schedule things - which line sends to
which other line during a cell slot.  Things get more complicated if
you have to send a cell to more than one other line.  The CBR case turned
out to be easy.  We didn't have any good ideas for the ABR case and
we didn't think it was important so we didn't worry about it.

So the software for that case has to process each cell - sending it to all
the outbound VCs one at a time.
[posted by Notes-News gateway]