---- Eben King <eben01@verizon.net> wrote:
> From Wayne Pollock:
>
> ---------- Forwarded message ----------
> Finally, while it is possible to violate standards by
> using longer cables, the 100 meter limit per segment
> still applies. It takes time for electricity to
> propagate through a wire, and Ethernet devices will
> only wait so long before assuming the segment isn't
> in use and begin transmission of their own. Remember
> you're using a shared media with Ethernet, so a too
> long segment will result in intermittent dropped
> packets. (Of course using full duplex, switched
> Ethernet may mitigate that and allow longer segments.)
>
> -Wayne
To add to this a bit, ethernet is officially known as CSMA/CD.
Carrier Sense with Multiple Access/Collision Detect. This
just means that every station has to "fight" for access to a
shared bus (Multiple Access) and has to incorporate a
mechanism to tell when more than one station is trying
to access the backbone buss (Collision Detect).
The protocol was designed so that the first signal in every
ethernet packet is the Preamble. This is an alternating
binary 0 and 1 bit stream, intended for all other receivers
on the bus to tell that the buss is occupied, with no
regard to possible data loss.
Each local LAN is a "collision domain" (crossing a repeater
or router means your signal leaves the CD, or, collision domain).
The two factors that set an outside bound on any CD's size are
the propagation delay for a packet to cross the max. length
of the CD (time) and the attenuation of the wiring (signal level
becoming too weak to trigger another receiver).
The preamble length originally 56 bits was designed so that
the max size of the CD would enable any station to hear
a preamble from the furthrest other station, then hold off
its own sending, which would otherwise cause a collision. The
preamble is simply a binary 10101010 repeated, to let
the other transceivers sync up to it, then ended with a
10101011 (end delimiter), then followed by the dest. MAC
address. That's where they typical NIC starts paying
attention to the data content of the frame.
Only a specialized packet analyzer such as a Sniffer (TM)
or a Wandel & Goltermann, or a Shomiti, would decode
fragments of a preamble that might be present if you have
some broken hardware on the LAN. These show us as
12 to 30 (or so) bytes of hex AA AA or 55 55, which are
composed of 01010101 or 10101010. This can really
slow down your net throughput and most packet analyzers
won't even reveal the presence of such junk. This is all
strictly PHY (Layer 1 stuff) or, just meaningless voltage
on the wire, with no inherent value in the "data". The
newer gear is less susceptible to this problem, but bad
wiring can lead to it, not to mention the crosstalk issue.
When a transceiver in a NIC becomes deaf, it will not
hear the preamble, so may start sending "blind" while
thinking the buss is clear, This guarantees a re-transmission
by the other station and is known as a Late Collision.
The Binary Exponential Backoff was developed so that
two stations, both trying to send, would not backoff identically
then re-send again at the same time. The specified limit of
no more than 1024 stations on any single LAN is a representation
of fairness given to any one station that could theoretically be
forced to back off the max. possible (8x) (due to collisions)
and yet still be afforded a chance to eventually transmit.
Collisions are a normal part of ethernet and are expected, to
a reasonable value that does not unduly affect throughput.
If your app needs a more guaranteed delivery time, you
might need a deterministic format. Token Ring has zero
collisions and a max. size (length/time) for a token rotation,
but I doubt anyone specs TR nowadays as ethernet works
well enough. Also the reassembly feature of TCP is in
itself, as well as the MAC level checksumming, a
guarantee against data loss.
- Bob Foxworth (a former Wayne Pollock student)
(who does packet analysis for fun, if not profit...)
-----------------------------------------------------------------------
This list is provided as an unmoderated internet service by Networked
Knowledge Systems (NKS). Views and opinions expressed in messages
posted are those of the author and do not necessarily reflect the
official policy or position of NKS or any of its employees.
This archive was generated by hypermail 2.1.3 : Fri Aug 01 2014 - 16:06:14 EDT