Path: utzoo!attcan!utgpu!jarvis.csri.toronto.edu!mailrus!iuvax!purdue!bu-cs!kwe
From: kwe@bu-cs.BU.EDU (kwe@bu-it.bu.edu (Kent W. England))
Newsgroups: comp.dcom.lans
Subject: Re: Ethernet bridge recommendations?
Summary: Keep in mind the diff between filtering and forwarding
Message-ID: <34849@bu-cs.BU.EDU>
Date: 18 Jul 89 16:46:16 GMT
References: <2926@sharkey.cc.umich.edu> <3769@ursa-major.SPDCC.COM>
Reply-To: kwe@buit13.bu.edu (Kent England)
Followup-To: comp.dcom.lans
Organization: Boston U. Information Technology
Lines: 36

In article <3769@ursa-major.SPDCC.COM> eli@ursa-major.spdcc.COM
> (Steve Elias) writes:
>>I have about $6000 (at most $8000) to bridge 3 standard ethernet subnets,
>>so far the best deal I can find is on 2 of Retix' low-end bridges that
>>are rated for 6000 'frames/second'... is that fast enough for joining segments
>>that have about 10 hosts each?? 
>
>	if the bridges can maintain 6k frames/sec steady state, then
>	they should be plenty fast.  unfortunately, many bridge companies
>	are playing bullshit specsmanship games.  example:  quoting
>	"burst" performance numbers.  taken to the limit, one could quote
>	infinite performance using this technique.  beware.
>
	Keep in mind the difference between filtering and forwarding
rates.  A bridge has to be able to scan each and every frame that
flies by.  This is the filtering rate.  It should be greater than your
expected Ethernet segment traffic.  Make sure you understand whether
it is total or per interface.  You might have to take the quoted
performance and divide by two.

	The forwarding rate is how fast it can move frames that need
to be sent to the "other side".  Burst rate is important, as for
example, in a network file system access, where a server might blast
several frames constituting some large chunk of data at the bridge
back to back (or nearly so).  Steady state throughput is another
important measure of how much offnet traffic from all sources can be
forwarded.

	Then you need to know whether you need or care about spanning
tree, learning, broadcast forwarding control and extra filtering for
access and security.

	Steve is right about the specsmanship.  This is unfortunately
something we have to put up with until we can decide just what we mean
by bridge and router performance and how it is we measure traffic on a
net.