Date: Sun, 22 Aug 1999 19:56:54 -0600 From: Wes Peters <wes@softweyr.com> To: "Rodney W. Grimes" <freebsd@gndrsh.dnsmgr.net> Cc: Chris Dillon <cdillon@wolves.k12.mo.us>, Cliff Skolnick <cliff@steam.com>, jay d <service_account@yahoo.com>, Evren Yurtesen <yurtesen@ispro.net.tr>, freebsd-security@FreeBSD.ORG Subject: Re: multiple machines in the same network Message-ID: <37C0AA66.FD70C73C@softweyr.com> References: <199908220649.XAA31700@gndrsh.dnsmgr.net>
next in thread | previous in thread | raw e-mail | index | archive | help
"Rodney W. Grimes" wrote:
>
> Your making a common mistake here when an ``ALL PORTS FULL LOAD'' test
> is done, if you have 40 ports all being sent data at 100MB/sec that
> data is going to have to come out on 40 ports someplace, so you only need
> 4Gbit/sec of backplane to do this. Thats 4G bytes of data in, 4G
> accross the backplane, and 4G back out of the box.
>
> Maybe a drawing would help:
>
> rxpair of port 1 > +---------+ > txpair of port n
> rxpair of port 2 > | | ....
> rxpair of port 3 > | Fabric | > txpair of port 3
> ... | | > txpair of port 2
> rxpair of port n > +---------+ > txpair of port 1
>
> As you can see the Fabric only has to handle 40 x 100Mb/s to
> keep all 40 ports busy at full duplex.
>
> The 3.8 Gb/s spec comes up a little short, but only buy 2 ports...
> and it had better be darned efficent as far as overhead goes...
>
> Allowing the port cards to short circuit bridge (and every switch
> chip set I have looked at does this) makes it easy to pass this
> test, infact you can do it with 0 load on the backplane. My
> drawing above tends to put the maximal load on a switches backplane,
> but unless the vendor tells you exactly how they tested the benchmark
> is like any other benchmark without all the nitty gritty details,
> total sales and marketing propoganda.
That's what I said to Chris, only I said it a little nicer. I have to,
since I'm part of that industry. ;^) To paraphrase Mark Twain: "There
are 3 kinds of lies: lies, damned lies, and benchmarks."
On the OS4024 we were discussing, it wouldn't matter which port pairs
you picked, because they're all on the same NI - the 4024 is basically
a single Network Interface built into a shruken chassis. If the traffic
is all in the same VLAN, it will only get to the "backplane" once, to the
source learning process. After that, the packets will always be switched
on the NI. Performance then becomes a matter of efficient buffering.
--
"Where am I, and what am I doing in this handbasket?"
Wes Peters Softweyr LLC
http://softweyr.com/ wes@softweyr.com
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-security" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?37C0AA66.FD70C73C>