[3494] in linux-net channel archive
Re: shaper or whatever
daemon@ATHENA.MIT.EDU (Dennis)
Tue Jun 25 16:49:03 1996
Date: Tue, 25 Jun 1996 11:21:17 -0400
To: Alan Cox <alan@cymru.net>
From: dennis@etinc.com (Dennis)
Cc: linux-net@vger.rutgers.edu
>> What we're doing is speed simulation so ISPs can do web and host
>> serving on an ethernet and charge for bandwidth usage, or so that
>> companies can share an ethernet and allocate only certain amounts
>> of bandwidth to each "user". For example if you had a T1 and you
>> wanted to put the company next door on you LAN but they only needed
>> 56kbs, rather than putting up back-to-back routers you can just run
>> an ethernet and limit the bandwidth on the ethernet. Easier, cheaper
>> and more efficient (hopefully).
>
>Its only efficient if your model is ok. The question thats critical is "What
>throughput do I get over a 56K link and do I get the same throughput over my
>56K limited ethernet". Here throughput is not bits/second its transactions,
>ftp data rate - tangible performance items. If the traffic shaper means I
>still get 56Kbit put I get 2/3rd of the ftp performance of a real 56K line
>then you can forget it.
The model needs to be better than "OK" :-)
The bits per second IS the critera, because
if you dont toss any packets then the performance of the application has to
be the
same. Remember that Ethernet headers are bigger than PPP headers, so there
will be a miniscule loss there...but you can compensate by using 56400 or
something
if you want to be really anal about it.
Well, I ran it all weekend with continuous data taking performance samples
every 60 seconds. Every sample was between 55900 and 56100. With virtually
no load. Not bad, I'd say. Ready for release!
FTP performance is like 6.8k or so (but I hate using it because there are so
many
variences....two tests that yield even slightly different numbers in
identical conditions
bother me alot.
If you get 2/3 the performance, then your model is not only not "OK", it is
defective.
Dennis
----------------------------------------------------------------------------
Emerging Technologies, Inc. http://www.etinc.com
Synchronous Communications Cards and Routers For
Discriminating Tastes. 56k to T1 and beyond. Frame
Relay, PPP, HDLC, and X.25 for BSD/OS, FreeBSD
and LINUX
