Conference 7.286::fddi

1023.0. "Good Publicity for DEC's EISA FDDI Adapter! Gets "thumbs up" from Network Computing" by QUIVER::WASHABAUGH (Born to be Mild) Sat Jul 10 1993 14:28

In article <[email protected]>, [email protected] (Art Wittmann) writes:
Newsgroups: comp.dcom.lans.fddi
Path: nntpd.lkg.dec.com!nntpd2.cxo.dec.com!pa.dec.com!decwrl!olivea!spool.mu.edu!howland.reston.ans.net!darwin.sura.net!news-feed-1.peachnet.edu!umn.edu!doug.cae.wisc.edu!wittmann
From: [email protected] (Art Wittmann)
Subject: FDDI EISA adapter test results
Organization: College of Engineering, Univ. of Wisconsin-Madison
Date: 9 Jul 93 13:42:26 CDT
Message-ID: <[email protected]>
Originator: [email protected]
Lines: 71


I got such a huge number of requests for these stats that I decided
to just post them here, rather than mail them individually.

We tested five different EISA cards, these where: the 3COM 3C770,
the DEC DEFEA-AA, the Network Peripherals EISA-3, the Crescendo CRS320
and the SysKonnect SK-5341.

We wanted to test these cards in the ways that we thought they'd be
used.  To that end we did two different tests; both involved using the
EISA cards in Novell servers and pumping out IPX packets.  Our ring
here at the UW college of Engineering is made up of 5 cisco AGS+4 routers 
and crescendo concentrators with various servers on the concentrators.
We felt that the most likely scenerio for use is to put servers on a 
ring and leave most stations on ethernet (or token ring) with routers
in between.  We also felt that some stations would go directly onto
FDDI so we tested both ways.

Our test server was an HP Vectra 486/33T with 32 Mbytes of memory.  
The test clients where HP 486s of various types, most had NE2000 
Ethernet cards.  All clients where running DOS 5 and Novell's latest
shells.

We wanted to make certain that the cisco's where not the bottleneck,
so we used stations that where on five different ethernet segments 
connected to three different routers.  We wrote a program that controlled
these different stations through in-band signalling.  To measure the
actual throughput, we used a Tekelec Chamelan 100 FDDI analyzer.  The
program generated traffic by reading cached data from the server.

The numbers below are Mbits/second / server % utilization.

FDDI to Ethernet:

Card		4 stations	7 stations 	20 stations	34 Stations
3com		6.0 / 33%	  9.6 / 50%	 14.4 / 76%	  17.5 / 85%
DEC		6.8 / 46%	  8.8 / 53%	 12.8 / 80%	  16.8 / 89%
Crescendo	5.6 / 36%	  9.6 / 55%	 13.6 / 67%	  16.8 / 85%
NP		7.6 / 46%	  9.6 / 56%	 13.6 / 73%	  17.8 / 90%
SysKonnect	7.2 / 33%	  10.4 / 55%	 14.4 / 72%	  17.6 / 85%

FDDI to FDDI:
Card		1 Station		2 Stations
3com		12 / 23%		21.6 / 50%
DEC		14.4 / 20%		27.2 / 40%
NP		13.6 / 23%		27.0 / 44%
Crescendo	9.6 / 28%		19.2 / 54%
SysKonnect	14.4 / 23%		26.4 / 46%

Remember, your milage may vary (blah, blah, blah).

My conclusion from the above was that in the FDDI to Ethernet test,
the FDDI cards themselves where not the bottleneck, rather it was the
server CPU.  If you are going to use cards in servers that way, performance
of the card is probably not the critical criteria.

The FDDI to FDDI test, the cards to make a big difference.  In the 
article, we gave our editorial okeedokee to SK, NP and DEC.  Different
drivers could probably significantly change the performance of any of these
cards.

Now remember, you still have to read this article when it appears in Network
Computing and you have to buy all products advertised in NWC!

Art

=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Art Wittmann                                  Phone: (608) 263-1748
Associate Director                            Email: [email protected]
Computer Aided Engineering Center                or: [email protected]
University of Wisconsin, Madison