[comp.protocols.tcp-ip] Request of info on bridges or gateways

kwe@bu-cs.BU.EDU (kwe@bu-it.bu.edu (Kent W. England)) (02/10/89)

In article <8902080638.AA14652@SCUBED.ARPA> 
>hutton@scubed.arpa (Thomas Hutton) writes:
>We are soon going to need to connect our three San Diego sites ethernets
>via T1 lines.  I am interseted in hearing recomendations on various
>hardware for doing this.  
>
>Since we need to transfer not only IP traffic but DECNET as well Im assuming
>I need to use a bridge vrs a gateway product.  One of the things we would
>like to do is have a redundent set of circuits:
>
	If you install three bridges, you will need spanning tree or
packet loops will result.  You will not be able to use any part of the
bandwidth of the redundant link.

	If you install routers, you can use all three links for
throughput and for redundancy.  If you need IP and DECnet, you can
find routers that will route both.  If you want bridging too (like
LAT, ugh) you can buy a cisco HyBridge which will route IP and DECnet
and bridge whatever other protocols you want.  I have not yet
personally set up a HyBridge, but I read the Rel 7.0 addendum and it
looks like a reasonable hybrid.

	The new cisco hardware is full-ethernet bandwidth, so the old
argument about packet switching slowness of routers is now moot.
>
>I am concerned with the problem of routing loops.  I would really
> like something
>that could while avoiding loop problems, 
>split the traffic over the links
>to optimize the bandwidth of the links.
>
	All modern routing implementations should have no trouble with
three links in your configuration.  Spanning tree will stop bridge
loops, but lose bandwidth, idling one link completely.

	BTW, are your sites within 4-5 miles line of sight of each
other?  You could save a lot of money with microwave and you could run
full Ethernet bandwidth, too.  There are, at last count, three vendors
selling such products.

	Kent England, Boston University