[comp.dcom.lans] E-NET REPEATERS

bobkil@ibmpcug.co.uk (Bob Kilgore) (02/19/91)

In response to the repeater postings, the repeater rules are as follows:

A network (collision domain) may have as many repeaters as is required to
build the network.

There may be no more than four (4) repeaters in the signal path between any
two (2) nodes in the network.

A multi-port repeater acts as one repeater in the path.

The devices that vendors often call 'hubs', or 'concentrators' in reference
to 10BASE-T segments are repeaters as far as the network is concerned.

The following diagrams are representative of conformant networks.

Due to collision timing a conformant network can have five (5) segments of
which two (2) must be link segments with no more than two devices attached.
These link segments may be fibre-optic, coax, or twisted pair.

Sn  = regular segment with nodes.
LSn = Link Segments with only end devices.

  node-----(rptr)-----(rptr)-----(rptr)-----(rptr)-----node
       S1         LS1        S2         LS2        S3

MPR=Multi-port repeater.

                          (MPR)
         port 1       -----------     port 6
     _________________| | | | | |____________________
     |        __________| | | |____________          |
     |        |        ___| |____          |         |
     |        |        |         |         |         |
   (MPR)    (MPR)    (MPR)     (MPR)     (MPR)     (MPR)
     |        |        |         |         |         |
  n1--n15  n1--n15  n1--n15   n1--n15   n1--n15   n1--n15

The above has 90 nodes and 7 repeaters for a 10BASE-T net.  In the signal
path between any two nodes is 3 repeaters, therefor this is a conformant
network.  The above example is typically how 10BASE-T networks are
designed.  The max. nodes with all 16 port MPR's is 240, with all 32 port
MPR's 992.  It is easy to see that a max node network (collision domain),
1024, is possible and very simple.

You will also note that this diagram fits very nicely into the now popular
'structured wiring concept' as brought to you by AT&T. By using a common
backplane and a variety of module cards it is possible to connect any media
to any other media so long as all the segments are 10BASE.

The reason for the limit of four (4) repeaters is only partly dependent on
delay, as some suspect.  As Phil Blanchford pointed out this delay can be
compensated by shortening the maximum distance from one end node to the far
end node.  All repeaters must insert enough bits into the preamble to
insure at least 56 bits of preamble are transmitted.  Further, repeaters
are allowed to insert bits into the preamble to insure their FIFO is filled
and clock skew is compensated.

In networks where the media 'eats' some number of bits the repeater will
restore the preamble.  As media got better, transceiver quality was
improved and new media were introduced, that didn't 'eat' any bits, another
problem was encountered. As each repeater may adds bits to the preamble,
the segment at the far end of a network has an 'extended' preamble.  It can
be shown that the effect of this is to shorten the time between the last
transmitted bit of packet A, and the first bit of packet B if both packets
originated from the furthest segment. This phenomenon is known as 'Inter
Packet Gap' shrinkage, and is referred to in the 10BASE-T document.  The
analysis of IPG is very complex and involves multiple variables. These
include the variability in bit loss for a given packet in the MAU's and
repeaters.  This could shorten the IPG to the extent that a MAU may not be
able to recover and receive the next packet.

The design of a network is generally a set of compromises as to what is
possible and what the client wants.  It is possible to design a network
that will function as a conformant Ethernet, but have a minor flaw with
respect to the standard.  Remember that the standard was written for the
worst possible case.  I have seen a document that was published in a noted
journal that measured the actual possible throughput of an Ethernet, by a
well known author that was employed by a well known Ethernet company, that
had 5 repeaters in the signal path.  The result of 'breaking a rule' are
not always catastrophic.  The trick for the network designer is to know
when and by how much.

When contemplating network design it is wise to consider the future.
Expansion of a network is almost inevitable and if you have designed a
nonconforming network the expansion could have catastrophic results.  With
careful design, 4 repeaters in the signal path are adequate for 8802-3
networks.  Avoid nonconforming networks, but know what the limits are, and
why they were imposed.

BOB
bobkil@ibmpcug.co.uk
-- 
Automatic Disclaimer:
The views expressed above are those of the author alone and may not
represent the views of the IBM PC User Group.
--