zeleznik@cs.utah.edu (Mike Zeleznik) (05/11/91)
This is a summary of the little I found out regarding an earlier request
for info on physical Ethernet testing. Thanks to those who responded.
-----------------------------------------------------------------------
Synopsis of original posting (from November 1990)
I am looking for a device to analyze physical signals on an Ethernet
(e.g., fall time, jitter, dc/ac component), while being able to select
the signals of interest based on higher level protocol information
(like a protocol analyzer).
For example, I may suspect that signals from a station at one end of a
cable segment are badly degraded by the time they reach the other end.
Since I usually do not have the option of isolating the network, I
would like to hang a probe on the net, filter out only packets with
appropriate src/dest addresses, and then obtain information about the
actual physical signals associated with those packets.
Of course, I can hang a scope on the cable, but I can not easily
display only the packets of interest. One conceptual solution is a
protocol analyzer with a trigger output, tied into a digital scope.
We would also like to analyze signals on the AUI cables. We had a
situation where a Sun Sparcstation, a Cabletron Multiport Transceiver,
and an HP MAU, would not work together. The Sun would not see some
packets destined for it, but a Sniffer attached to another port on the
multiport would see them all. Replacing the HP MAU with a Cabletron
MAU fixed the problem! We have since similar situations with other
combinations of equipment. It would be nice to see what is going on
at the AUI level.
-----------------------------------------------------------------------
Summary of available tools:
I received very few replies. The following is what I know to date.
They are discussed below.
Network Quality Analyzer (NQA).
NETCOM SYSTEMS Ethernet Tester/Simulator/Analyzer.
Network General Sniffer trigger output fed to digital storage scope.
HP and Tektronix (especially their LP COM Division) have a wide
range of testing devices, but they do not seem to fit our needs.
-----------------------------------------------------------------------
Network Quality Analyzer (NQA)
Made by Logic Replacement Technology Ltd., distributed by American
Spectrum, 15261 Barranca Parkway, Irvine, California 92718, (714)
727-3944.
NQA appears to do much of what we want, such as measuring jitter,
AC/DC components, and fall time. It also appears to present this
information in a useful way. But it seems to be limited to the
ethernet cable only, and is expensive (on the order of $30K). You are
paying for a full protocol analyzer plus the physical end.
The following is from an earlier net posting (June 1990), which seems
to accurately describe what I understand the product to do:
Begin: previous posting: --------------------
From:
Charles Spurgeon | |
UTnet Network Information Center | spurgeon@emx.utexas.edu |
University of Texas at Austin | ...!cs.utexas.edu!ut-emx!spurgeon|
The NQA uses a touch sensitive screen to walk through a menu-based
system that includes several level-one physical network mediuma
measurements. It has an accurate DTDR (digital time domain
reflectometer), builds a jitter graph for the devices on your network,
and will also look at the A.C. and D.C. signal components, bit rate,
and fall time for each station. ... Here's a brief look at a few of
the tests:
The jitter graph displays a set of small numbered graphs, with a
larger window next to them showing the association between graph
number and Ethernet station address. Stations that produce excessive
signal jitter are highlighted.
The bit rate display shows the accuracy of the clock on a given
Ethernet controller in terms of whether the bit rate is inside or
outside the apercentage of error allowed in the 802.3 specs. The NQA
displays a bar graph for each controller on the wire. You could get a
feel for which controllers were pushing the bit rate, and after some
time you could begin to identify certain vendors by the bit rate graph
on their interface. "That's an xyz interface, their bit rate is
always right on the edge."
The NQA also checks the coax for DC bias voltage, and will look at the
DC component of a packet "as recovered by a 2 usec (approx) low pass
filter. This value is measured as a voltage but presented as a
current, since co-ax impedance is known. The IEEE specification for
the parameter is quoted as a current." "The collision detect
threshold is a value built into every transceiver on the network and
hence is subject to a manufacturing spread. The DC component launched
by a transceiver is also subject to a manufacturing spread and so
there is some scope for conflict in which some transceivers see
spurious collisions from certain stations but not other."
The DTDR is quite nice, works on a live network, and lets you zoom in
and out on the DTDR display. You can save DTDR images to disk.
Working with this device was like having a microscope for the signals
on an Ethernet coax. The NQA also does level-two measurements such as
a traffic matrix, traffic analysis, alarms on packet events, and so
on. Remote "pods" can be attached to coax segments linked via bridges
or repeaters, and the NQA device will display the signal levels on the
remote segments as well.
I could go on, but you get the idea. Watching the NQA in action is
impressive. The price is impressive as well - on the order of $26,000
US, and remote pods cost another $7K or so. Although to be fair, high
end protocol analyzers from Network General and HP cost this much too,
and I haven't seen anything else on the market that can perform the
coax electrical measurements that the NQA does. I suppose it might be
possible to lease one of these things for a while, to see how bad your
network is.
End: previous posting: --------------------
-----------------------------------------------------------------------
NETCOM SYSTEMS Ethernet Tester/Simulator/Analyzer:
NETCOM SYSTEMS Inc. 21828 Lassen St., Unit G, Chatsworth, CA 91311,
(818) 700-0111.
They make a box called the ET-100, an 802.3 Ethernet tester/simulator/
analyzer that is rather impressive; but I am not sure it will really
help in the situation of interest here.
It provides two AUI interfaces, with the ability to test things such
as errors, SQE, collision circuitry, jabber timer. It can generate
network loads at intervals from 1 sec down to 9.6 micro-sec, can force
long/short collisions; basically just about anything you want to test.
While it seems excellent for proving whether a particular device
works, or how well it works (e.g., bridge forwarding rate), I do not
see how much it will tell us about the problem we are interested in
(we already know the device combination does not work; the question is
why).
They offer a rental policy, deductible from purchase, so it might be
worth checking out down the line. Price is around $6,700.
-----------------------------------------------------------------------
Feeding trigger output from Sniffer to digital scope:
The idea here is to use the Sniffer to select the packet of interest,
generating a trigger signal which causes the scope to stop capturing
data.
Version 3.0 of the Sniffer software does provide a trigger output
signal on the serial port. I do not know if other protocol analyzers
can provide this (e.g., the Lanalyzer does not appear to).
I have not had the time to actually try this approach, or investigate
it in much depth. One concern is the time delay in getting the
trigger out of the Sniffer. If this is longer than the storage
capability of the scope, the scope may overwrite the data before the
trigger arrives. Another concern is how stable this delay is. If the
jitter is too great, it may be hard to know which packet caused it.
-----------------------------------------------------------------------
If anyone has additional information, I would like to hear it. More
when I know it...
Michael Zeleznik Computer Science Dept.
University of Utah
zeleznik@cs.utah.edu Salt Lake City, UT 84112
(801) 581-5617wsmith@cs.umn.edu (Warren Smith [Randy]) (05/11/91)
If you want a really CHEAP (in both senses) method of testing signal quality you can use an oscilloscope and a simple packet generator running on a workstation. Assuming you have a scope handy the cost is nothing. By having one station blast the packets out as fast as it can and triggering off the start of the packet you get a pretty good idea what the packets look like. We've used this technique to get a look at signal quality. I wrote a program on a Sun to blast out packets and modify the data bits every 100 packets or so. We could see this pretty well with a 40 MHz scope, however I think you probably need a 100 MHz scope to get a really good look at the signal. We could see the signal deteriorate as we increased the length of the run (can't remember what we were testing though - maybe twisted pair?). Its no NQA, but then what is? -Randy -- Randy Smith wsmith@cs.umn.edu ...!rutgers!umn-cs!wsmith