frank@CAEN.ENGIN.UMICH.EDU (Randy Frank) (11/17/90)
For about two months we have been beta-testing an implementation of
Proteon's routing software for Apollo Domain protocol type (DDS). Our
understanding from Proteon is that production release of this software
is slated for mid-December, although you should probably check with
Proteon directly for more information and pricing.
Proteon has done this project cooperatively with Apollo, and uses and
off-the-shelf Apollo PC/AT ring cards in their routers for connecting
to the ring. Proteon also supports DDS routing on the 4200, which we have
and plan on using to connect a remote ring (via ethernet) to the Proteon
80MB fiber backbone as a way of replacing an in-place T1 link on DSP90s.
However, we are not ready to report on this at this time as we haven't
completed testing. Proteon DOES NOT support direct connection of
Apollo rings to the 4200 (and therefore to Pronet 80), so you must go
thru either an Apollo or P4100+ from the ring onto an ethernet and then
to a 4200/Pronet 80.
The Proteon implementation supports both DDS routing (for example
between two rings), as well as IP gatewaying (for example between
a ring and a ethernet). We have tested both of these. They also
support DDS routing between a ring and an ethernet (for those with
Ethernet based Apollo workstations); however, we do not have any
Apollos on ethernet so we have not tested this.
We have tested and done extensive performance mesaurement of a Proteon
4100+ router in both ring-ring DDS routing and ring-ethernet IP routing.
For ring-ring DDS routing we have compared against DN3500s, DN4000, and
DN4500s with two ring cards (all of which we've used from time to time
as ring-ring routers.) For ring-ethernet IP routing we've compared
against DN3500 with a ring and ethernet card.
The main reason this may be of interest to large Apollo network users is
that in our tests the performance of the 4100+ router is substantially
better than all Apollo routers tested. (We would love to test a DN10000
as a router, but we don't have one configured with two network interfaces!
Also, shortly (Motorola not withstanding) you will be able to use a DN5500
68040 based system as an Apollo router, and we are obviously unable to
predict what the performance of this system as a router might be.)
We think that for people with heavy network loads use of the Proteon
P4100+ may be worth investigating. This may be particularly true (as in
our case) where we are using dedicated 3500/4500s as routers (i.e.,
no users are allowed to login to them). In this environment you may be
able to justify purchase of the 4100+ in terms of freeing up Apollo
nodes for the purpose for which they are intended. (Please don't ask
me about pricing - talk directly to Proteon.)
Once again, we are sending out this information to the Apollo list because
we think it may be of general interest. If you have any particular
questions about configuration of Proteon routers, costs, delivery, etc.,
please contact Proteon DIRECTLY.
Following is the measured performance data done by a member of the CAEN
staff who has been actually testing and measuring the performance of these
routers.
Randy Frank
-------------
Date: Sun, 4 Nov 90 22:46:18 EST
From: ahoover (Andrew P Hoover)
Message-Id: <4dd136e4d.000b617@caen.engin.umich.edu>
Subject: Proteon p4100 stats
Here are the results of this weekend's performance testing
of the Proteon p4100:
Router Type of Traffic Interface #1 Interface #2 max packets/second
mean packet size (bytes)
- ------ --------------- ------------ ------------
------------------ ------------------------
DN4000 DDS ring ring 350 +/- 10 570 +/- 10
DN4500 DDS ring ring 465 +/- 10 570 +/- 10
p4100 DDS ring ring 650 +/- 10 570 +/- 10
DN3500 DDS ring ethernet 280 +/- 10 570 +/- 10
p4100 DDS ring ethernet 440 +/- 10 570 +/- 10
p4100 DDS (see note 1) ring ethernet 470 +/- 10 570 +/- 10
DSP90 DDS ring IIC (T1) 153 +/- 5 570 +/- 10
DN3500 IP ring ethernet 170 +/- 10 980 +/- 10
p4100 IP ring ethernet 565 +/- 10 980 +/- 10
Note 1: Note that there are 2 sets of statistics for the p4100 routing
DDS to ethernet.
440 packets/second was achieved on a network that already had a
15% utilization.
470 packets/second was achieved on a network that conatined nothing but the
ring<-->ethernet routers. The statistics for the DN3500 DDS
ring<-->ethernet
are for an ethernet with a background load of 15%. All IP
testing was done on
ethernets w/ <2% background load.
Note the incredible speed improvement of routing IP across a p4100
rather than a DN3500 (230%
speed improvement == 3.3x the speed of a DN3500).
I don't intend to test ring<-->ring IP performance on the p4100 as I am quite
certain we will see similar improvements over our current router (//alonzo) and
because results from ring<-->ring IP testing are much more difficult to obtain
accurately.
I have yet to test DDS routing performance through a p4200's p80 and ethernet
interfaces. Once we have this data, we should have a pretty good idea of
the type of speed improvement we'll see from switching to Proteon routers
for the Ugli (T1) link. As it stands now, I predict we'll see 200-300% speed
improvement to Ugli if we get rid of the T1 and use p4100's on north campus
and central campus. This speed improvement drops to a 50-70% increase if
we get rid of the T1, but still use an apollo to go to ethernet in the Ugli.
Regardless of what we use in Ugli, it is not worth it to have a second ethernet
in the Ugli just for DDS.
--Andy
------- End of Forwarded Message