Ascend Archive
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: (ASCEND) I need some help.

On Tue, 14 Jul 1998 16:47:23 +0200, Andre Beck <beck@ibh.de> wrote:

> On Fri, Jul 10, 1998 at 09:06:52AM -0400, Jim Howard wrote:
> > At 20:04 1998/07/09 -0400, TJ Schuler wrote:
> > >The remote site has around 10 computers connected to a hub along with
> > >the 50. The local site has another 50 connected to one of our switches.
> > 
> > I've never tried bridging off of a switched port, but this might
> > be the source of some of your problems.
> 
> Shouldn't be any problem. Switches are essentially nothing but performance
> enhanced multiport bridges. 

"Essentially" is relative and might be misleading in this context.
Switches are _not_
multiport bridges.

1. If a packet with an unknown MAC address is received by a bridge, it
gets broadcast on _all_
   bridge ports. In a switch, it will only be broadcast onto the
backbone port (a port specifically 
   designated to receive unknown packets). The point here is that all
hosts on the other switch ports are
   known to the switch ("learned") - no need to waste bandwidth. A
switch does not learn addresses on the
   backbone port.
   Just to be cautious, there might be switches that are not
configurable in this way...
2. A further difference (but not relevant to this problem): if
collisions occur, a bridge drops packets. A switch
   might apply backpressure on the sending port, i.e. signal on layer 2
that the receiving end cannot cope.
   This is done with much less overhead than if the application would
have to deal with dropped packets.

> Bridges do speak the Spanning Tree Protocol.
> If the pipe implements bridging correctly (it should, at least if Dial
> Broadcast=Yes) it would just enter the spanning tree, cause a topology
> change and then live comfortable together with the other L2 stuff. Just
> keep in mind that convergence in a nontrivial STP domain takes 20 to 30
> seconds.

3. Spanning Tree is barely used in a switched network these days, mainly
because of the slow convergence time
   (up to minutes). A switch can make use of the link detection instead,
sometimes called "resilient links" or 
   "redundant trunk".
   I am not sure whether Pipelines support STP at all.
4. Performance in regard to latency time is indeed some orders of
magnitude better with switches.

Regarding the problem at the root of this thread (very significant
subject line BTW) I could imagine that
it might be caused by the interaction between the switch address
learning table and the MAXs' arp proxy.
What does the switch ADB look like - any WAN hosts showing up? What is
the proxy setting on the MAX?


Best regards,

   Wolfgang

_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-
                         B  E  N  E  I  C  K  E
                               EDV-Beratung
________________________________________________________________________
        Netzwerk-Design - 3Com Solution Partner
           ISDN Remote Access - ASCEND Vertrieb
              DIGITAL Partner PC-Systeme und Server
                 ApplePoint
                    Windows NT-Netzwerke
                       Workstation Memories and Storage Solutions

Dr. Wolfgang Beneicke                                 fon +49-6223-48126
Fasanenstrasse 16, D-69251 Gaiberg                     fax +49-6223-5708
...near world famous Heidelberg, Germany
_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-

++ Ascend Users Mailing List ++
To unsubscribe:	send unsubscribe to ascend-users-request@bungi.com
To get FAQ'd:	<http://www.nealis.net/ascend/faq>
Follow-Ups: